Language Change And Linguistic Inquiry In A World Of Multicompetence.pdf

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/331717887

Language change and linguistic inquiry in a world of multicompetence: Sustained phonetic drift and its implications for behavioral linguistic research Article  in  Journal of Phonetics · March 2019 CITATIONS

READS

0

123

1 author: Charles B. Chang Boston University 53 PUBLICATIONS   409 CITATIONS    SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Tone Aptitude View project

Korean Fricatives in Diachronic Perspective View project

All content following this page was uploaded by Charles B. Chang on 13 March 2019.

The user has requested enhancement of the downloaded file.

Language change and linguistic inquiry in a world of multicompetence: Sustained phonetic drift and its implications for behavioral linguistic research Charles B. Changa a

Boston University, Department of Linguistics, 621 Commonwealth Avenue, Boston, MA 02215, USA

Abstract Linguistic studies focusing on monolinguals have often examined individuals with considerable experience using another language. Results of a methodological review suggest that conflating ostensibly ‘multicompetent’ individuals with monolinguals is still common practice. A year-long longitudinal study of speech production demonstrates why this practice is problematic. Adult native English speakers recently arrived in Korea showed significant changes in their production of English stops and vowels (in terms of voice onset time, fundamental frequency, and formant frequencies) during Korean classes and continued to show altered English production a year later, months after their last Korean class. Consistent with an I NCIDENTAL P ROCESSING H YPOTHESIS (IPH) concerning the processing of ambient linguistic input, some changes persisted even in speakers who reported limited active use of Korean in their daily life. These patterns thus suggest that the linguistic experience obtained in a foreign language environment induces and then prolongs restructuring of the native language, making the multicompetent native Email address: [email protected] (Charles B. Chang) URL: http://cbchang.com (Charles B. Chang)

Preprint submitted to Journal of Phonetics

March 11, 2019

speaker in a foreign language environment unrepresentative of a monolingual in a native language environment. Such restructuring supports the view that one’s native language continues to evolve in adulthood, highlighting the need for researchers to be explicit about a population under study and to accordingly control (and describe) language background in a study sample. Keywords: ambient exposure, first language attrition, plasticity, crosslinguistic influence, voice onset time, fundamental frequency, vowel formants

2

1

1. Introduction

2

A fundamental goal of linguistic inquiry has been to characterize the complex

3

and largely unconscious knowledge that permits human beings to speak their first

4

language like a native speaker, with little apparent effort. To make inroads on this

5

task, researchers have often abstracted away from a native speaker’s knowledge

6

of other languages, with the result that, especially since the advent of the genera-

7

tive grammar movement in the 1950s (Chomsky, 1957), linguistic competence has

8

been analyzed largely in accordance with a MONOLINGUAL MODEL of the native

9

speaker. This approach is supported by some findings suggesting that bilinguals

10

develop distinct systems for their two languages (Genesee, 1989; Kim et al., 1997;

11

Paradis, 2001; Freedman & Barlow, 2012). However, the monolingual model is

12

problematic for two reasons. First, it is inconsistent with the fact that the ma-

13

jority of language users across the world are not actually monolingual (Tucker,

14

1999). Second, there is abundant evidence that a bilingual’s language systems

15

are not completely separate, but rather shared to some degree (Schwanenflugel

16

& Rey, 1986; Fox, 1996; Marian et al., 2003; Flege, 2007). Since a bilingual

17

cannot be considered the sum total of two monolinguals (Grosjean, 1985, 1989),

18

the investigation of bilinguals’ competence in just one language without regard

19

for their competence in the other language amounts to a questionable enterprise.

20

Nevertheless, this remains common practice within the field of linguistics, and the

21

ramifications of this practice for the study of language is the topic of this article.

22

This paper has three main objectives. The first is to show that the practice

23

of conflating bilingual and monolingual individuals is indeed common in the lin-

24

guistic literature and is, therefore, an issue that needs to be addressed. The sec-

25

ond is to show why this issue does not pertain to fluent, ‘balanced’ bilinguals 3

26

specifically, but rather to bilinguals more generally, including marginally bilin-

27

gual individuals—those who use a second language (L2) much less proficiently

28

and/or frequently than their native or first language (L1). The final objective is to

29

discuss methods of addressing the matter of language background in behavioral

30

linguistic research so as to increase the rigor, transparency, and generalizability of

31

empirical findings.

32

The rest of the paper is divided into five sections. In §2, I establish the premise

33

that bilinguals differ from monolinguals by synthesizing the research on effects

34

of L2 knowledge on the L1, with special attention to phonetic and phonological

35

effects of a late-acquired L2. The argument that, in spite of these effects, osten-

36

sibly bilingual individuals are often conflated with monolinguals is developed in

37

§3, which presents a review of population sampling methodology in behavioral

38

linguistic research. The scope of monolingual-bilingual differences—in particu-

39

lar, whether they extend to bilinguals who show limited active use of the L2—is

40

examined in §4, which reports a longitudinal study of L1 production in an L2 en-

41

vironment demonstrating the phonetic plasticity of the L1 in adulthood. In §5, the

42

findings are contextualized within the broader study of lifespan linguistic develop-

43

ment, with recommendations for the treatment of language background. Finally,

44

§6 summarizes the main conclusions.

45

2. Background

46

2.1. L2 influence on the L1 at multiple levels

47

Over the previous decades, a growing body of evidence has suggested that the

48

L1 system can be influenced by L2 experience. Clearly, knowledge and use of an

49

L2 is associated with metalinguistic consequences (Yelland et al., 1993; Jessner, 4

50

1999; Bournot-Trites & Tellowitz, 2002) as well as domain-general effects (Cook,

51

1997; Bialystok & Craik, 2010; Bassetti & Cook, 2011; Kroll et al., 2014), but

52

the main concern here is with linguistic effects, which often arise from late L2

53

learning in a variety of linguistic domains (Pavlenko, 2000). For example, L2

54

influence is observed at the conceptual and cognitive linguistic levels, where it is

55

evident that neither advanced L2 proficiency nor L2 immersion is required for L2

56

knowledge to result in L1 modifications (Brown, 2008; Brown & Gullberg, 2011;

57

Brown & Gullberg, 2012; for recent reviews, see Jarvis & Pavlenko, 2008 and

58

Higby & Obler, 2014). L2 influence is amply documented at the morphosyntac-

59

tic level as well (Kecskes, 1998; Pavlenko & Jarvis, 2002; Jarvis, 2003; Tsimpli

60

et al., 2004), in certain cases only when the L2 experience is extensive (Dussias

61

& Sagarra, 2007) or early-acquired (Kim et al., 2010) but in other cases just with

62

ambient L2 exposure (Laufer & Baladzhaeva, 2015). Additionally, L2 influence

63

is reported in lexical semantics as well as lexical access and processing (Pavlenko

64

& Jarvis, 2002; van Hell & Dijkstra, 2002; Pavlenko, 2003; Schmid & Köpke,

65

2009), even after a relatively short period of L2 immersion (Linck et al., 2009).

66

Studies of L2 users have led to specific claims about temporal and linguistic

67

constraints on L2-to-L1 influence within the bilingual mind. In regard to temporal

68

constraints, it has been stated, for example, that “L2 users who have been exposed

69

to the L2 for 3 years or longer through intensive interaction in the target language

70

context may start exhibiting bidirectional transfer effects in their two languages”

71

(Pavlenko & Jarvis, 2002, p. 209), while “a L2 that is hardly mastered should not

72

have much influence on L1” (Major, 1992, p. 201); such statements suggest that

73

L2 influence is a phenomenon specific to advanced L2 users. As for linguistic

74

constraints, it has been hypothesized that “changes in L1 syntax will be restricted

5

75

to the interface with the conceptual/intentional cognitive systems” (Tsimpli et al.,

76

2004, p. 257), reflecting the larger idea that L2 knowledge affects ‘fuzzy’ aspects

77

of the L1 such as meaning rather than core structural properties such as syntactic

78

parameters. As discussed above, however, while these hypothesized constraints

79

on L2 influence are consistent with some findings, they are not fully supported by

80

this literature, which also includes cases of L2 influence in non-advanced L2 users

81

as well as in structural aspects of the L1 such as inflectional morphology (Jarvis,

82

2003) and phonemic contrast (de Leeuw et al., 2018).

83

Positing globally restrictive constraints on L2 influence is especially difficult

84

in light of the abundant—and, in many cases, rapid—L2 effects documented at

85

the level of the sound (Kartushina et al., 2016; Celata, in press; Chang, in press;

86

de Leeuw, in press). With respect to phonological rules and contrasts, relatively

87

extensive L2 experience (in English) is found to alter the production and/or per-

88

ception of final devoicing in L1 Russian (Dmitrieva et al., 2010), /h/-merger in

89

L1 Korean (Joh et al., 2010; Cho & Lee, 2016), and the light-dark lateral con-

90

trast in L1 Albanian (de Leeuw et al., 2018). At the phonetic level, L2 experience

91

influences various properties of L1 speech, such as voice onset time (VOT), fun-

92

damental frequency (f0 ), and the first, second, and third formants (F1 , F2 , F3 ).

93

For example, late-onset L2 immersion in English (where voiceless stops have

94

long-lag VOT) leads to lengthened VOTs in the short-lag voiceless stops of L1

95

French, and vice versa (Flege, 1987); in fact, this VOT shift has been reported

96

even in functional monolinguals with only ambient L2 exposure (Caramazza &

97

Yeni-Komshian, 1974; cf. Fowler et al., 2008). With respect to f0 , L2 experi-

98

ence in Greek influences peak f0 alignment in L1 Dutch (Mennen, 2004), while

99

L2 experience in English is correlated with higher onset f0 values following lenis

6

100

stops in L1 Korean (Yoon, 2015). As for vowel formants, early-onset L2 experi-

101

ence in Spanish is linked to lower F1 values in L1 Quichua vowels (Guion, 2003),

102

while late-onset L2 immersion in English is linked to higher F1 values in most L1

103

Dutch vowels (Mayr et al., 2012). Late-onset English immersion can also affect

104

the production of laterals and rhotics in L1 German, as indexed by F1 , F2 , and F3

105

(de Leeuw et al., 2013; Ulbrich & Ordin, 2014; Bergmann et al., 2016).

106

Direct evidence of L1 change due to L2 influence has also been provided by a

107

number of longitudinal studies. For instance, in a one-year longitudinal study, Oh

108

et al. (2011) show that L2 immersion in English results in increased F2 values for

109

some L1 Japanese vowels, although only in children and not in adults. Additional

110

longitudinal data come from the L2 training literature (e.g., Kartushina, 2015;

111

Kartushina et al., 2016), as well as a case study of an L1 Portuguese late learner

112

of English (Sancier & Fowler, 1997; for related research on Spanish-English bilin-

113

guals, see Tobin et al., 2017). In the latter study, VOT in both L1 and L2 voiceless

114

stops is found to be influenced by the VOT norms of the most recently experi-

115

enced ambient language; thus, short-lag Portuguese stops are produced with sig-

116

nificantly longer VOTs following a few months of immersion in English, an effect

117

that is perceptible to native Portuguese listeners. The fact that this speaker’s L1

118

production is detectably affected by recent L2 immersion despite her greater to-

119

tal experience in the L1 is attributed to three factors: crosslinguistic phonological

120

similarity (which leads to L1 sounds becoming perceptually linked to, and thus

121

influenced by, similar L2 sounds), a tendency toward imitation (even of L2 expo-

122

sure; see, e.g., Ward et al., 2009), and the recency effect on memory.

123

Recent L2 experience, however, has an effect that is modulated by learners’

124

prior familiarity with the L2, as shown in longitudinal work on L1 English learn-

7

125

ers of Korean (Chang, 2012, 2013; for related research on L1 Mandarin learners

126

of Korean, see Holliday, 2015). Korean is a language that, unlike English, has

127

a three-way stop laryngeal contrast distinguished in terms of VOT and f0 (Yoon,

128

2015; Bang et al., 2018). In Chang’s results, both of these properties in the L1

129

show evidence of change due to recent L2 experience (PHONETIC

130

the first five weeks of L2 instruction in an immersion environment. Drift is found

131

in the VOT of English voiceless stops (which lengthens in approximation to the

132

longer VOT of the perceptually similar Korean aspirated stops), onset f0 (which

133

increases due to the f0 elevation associated with Korean fortis and aspirated stops),

134

and mean F1 of the vowel system (which decreases due to the lower mean F1 of

135

the Korean vowel system). Notably, regardless of the acoustic property examined,

136

the magnitude of drift is found to be larger in inexperienced learners (true begin-

137

ners) than in experienced learners with prior exposure to Korean, suggesting that

138

phonetic drift due to L2 learning decreases over the course of L2 development.

139

Although there are individual differences in phonetic drift (see, e.g., Huffman &

140

Schuhmann, 2015), the same general phenomenon is found in foreign language

141

(i.e., non-immersion) contexts (Herd et al., 2015; Schuhmann & Huffman, 2015).

142

Furthermore, recent work on phonetic drift in perception has demonstrated that,

143

like L1 production, L1 perception can undergo rapid shifts during elementary L2

144

learning as well (Tice & Woodley, 2012; see also Namjoshi et al., 2015).

DRIFT )

within

145

Thus, while there is some evidence that L2 learners can pattern like monolin-

146

guals in their L1, the bulk of the literature suggests that L2 experience tends to

147

influence L1 performance, regardless of when the L2 was learned. At the level of

148

syntax and semantics, there is some indication that L2 influence may be strongest

149

with an early onset and/or high level of L2 experience. At the phonetic level, by

8

150

contrast, effects of recent L2 experience are commonly found in late L2 learn-

151

ers and may be strongest at low levels of prior L2 experience. The occurrence of

152

crosslinguistic phonetic interaction in late L2 learners is, in fact, predicted by three

153

core principles of the Speech Learning Model (SLM; Flege, 1995, 1996, 2002):

154

(1) that sound categories continue to develop over the lifespan; (2) that the sounds

155

of an L1 and L2 exist in a shared mental phonetic space; and (3) that ‘similar’

156

(as opposed to ‘new’) L2 sounds tend to undergo a perceptually-based, automatic

157

‘equivalence classification’ with L1 sounds, resulting in a merging of their pho-

158

netic properties. According to the SLM, equivalence classification of L2 sounds

159

with L1 sounds becomes more likely as L1 categories evolve over the course of

160

normal L1 development (Flege, 1995), so the probability of perceptual linkage

161

between L1 and L2 sounds increases with a late onset of L2 learning. Adult L2

162

learners are thus particularly subject to L2-to-L1 influence, because they have the

163

greatest tendency to link L2 sounds to L1 sounds rather than creating distinct L2

164

categories.

165

2.2. Multicompetence and the notion of ‘native speaker’

166

The broad susceptibility of the L1 to L2 influence in adulthood is consistent

167

with a view of linguistic knowledge as fluid and holistic, encapsulated in Cook’s

168

notion of

169

Papp, 2000). In a multicompetence framework, the acquisition of additional

170

languages is conceptualized not as mere accrual, but instead as restructuring of

171

knowledge, a process that changes the language user fundamentally. Thus, com-

172

pared to unicompetent (monolingual) language users, multicompetent users have

173

not only more knowledge (of L2, L3, etc.), but also different knowledge overall,

174

including of the L1. As such, it is unexceptional—and actually expected—for an

MULTICOMPETENCE

(Cook, 1991, 1992, 2003; see also Kecskes &

9

175

L1 to be perceived or produced differently by L2 users compared to monolinguals,

176

who represent a mental state prior to ‘initiation’ into L2 knowledge.

177

In contradicting the idea of language separation within the multilingual mind,

178

the multicompetence framework also problematizes the term

NATIVE SPEAKER ,

179

an ambiguous descriptor for a linguistic profile that may or may not correspond

180

to ‘monolingual’ (see, e.g., Beinhoff, 2008). As a model for L2 learners, na-

181

tive speakers are commonly thought of as individuals who have attained ‘full’

182

command of the target variety, the type of command that is often the object of

183

description in linguistic research. In practice, however, the native (qua the most

184

proficient) speakers of a language are rarely identified as such by proficiency mea-

185

sures, but rather by proxy measures (e.g., age/onset of learning) or by self-report,

186

which may be based on the same proxy measures (e.g., being exposed to the lan-

187

guage from birth).

188

Building a sample of native speakers via proxy measures, while expedient,

189

may not pick out the intended population of native users—those who have “special

190

control” and “insider knowledge” of the language, who “control its maintenance

191

and shape its direction” (Davies, 2003, p. 1)—because most proxy measures do

192

not account for the fact that language knowledge is dynamic and, consequently,

193

subject to change as well as loss (de Bot et al., 1991; Seliger & Vago, 1991; Stol-

194

berg & Münch, 2010; Schmid, 2013). That is to say, native-like command at one

195

point in time does not necessarily imply native-like command at the time of study.

196

For this reason, the target population in linguistic research is often monolingual

197

native speakers, since monolinguals should be exempt from the transformative

198

influences of L2 knowledge, and this monolingual model is the one typically as-

199

sumed in research on one language (see, e.g., Chomsky, 1986; Piller, 2002).

10

200

2.3. Monolingual studies recruiting monolinguals?

201

Despite the commonness of the monolingual model in linguistic research, the

202

prevalence of multicompetence raises two important questions for this model. The

203

first question is whether this model makes sense for the language under study.

204

If, for example, the language is not generally spoken by monolinguals, but by

205

multilinguals, is it reasonable to examine knowledge of that one language without

206

considering the other languages within the user’s linguistic repertoire (see, e.g.,

207

Lüpke & Storch, 2013)? This paper does not address this question, although it

208

should be noted that the answer to this question might very well be no, depending

209

on whether the empirical simplification imposed by the monolingual model stands

210

to produce misleading results (e.g., where a study sample is composed of users

211

whose profile of multicompetence is not typical for users of that language).

212

Assuming that the monolingual model does make sense for the language under

213

study (because it is in fact spoken largely by monolinguals), the second question

214

is whether the monolingual model is being applied appropriately. That is, does a

215

study of ‘native’ (qua monolingual) knowledge of a given language actually exam-

216

ine monolingual users? Discrepancies in the linguistic literature suggest that the

217

answer to this question may often be no. For example, the phonology of Swedish

218

has been described inconsistently as contrasting either voiced and voiceless as-

219

pirated stops (Helgason & Ringen, 2008) or voiceless unaspirated and voiceless

220

aspirated stops (Keating et al., 1983), which is attributable to the latter study’s

221

examination of speakers immersed in L2 English.1 In other words, the two stud1

Since ‘the recordings for the experiment of Keating et al. (1983) were made in the US (Keat-

ing, p.c.)’ (Helgason & Ringen, 2008, p. 620), most likely at UCLA, the speakers’ place of residence was presumably somewhere in southern California, although the paper does not specify

11

222

ies describe Swedish phonology differently because although they target the same

223

population (namely, ‘native’ Swedish speakers), one examines functional mono-

224

linguals in a Swedish-speaking environment, whereas the other examines multi-

225

competent speakers in an English-speaking environment.

226

This type of methodological disparity reflects the assumption of an unchang-

227

ing L1—that is, the idea that L1 users maintain the same L1 competence regard-

228

less of variation in language background and environment. Given that this as-

229

sumption is questionable (see §2.1), it should generally be rejected; however, if

230

it is rejected while the monolingual model is maintained, then recruiting the tar-

231

get demographic for a behavioral study of one language (i.e., monolinguals) re-

232

quires understanding language users not just in terms of their self-identified L1,

233

but in terms of their broader language background and environment. This is be-

234

cause these latter variables, which both affect the linguistic behavior on which the

235

study’s conclusions are based, cannot be presumed to be the same (in particular,

236

monolingual) across so-called ‘native speakers’.

237

To my knowledge, the extent to which the field of linguistics, including the

238

subfield of phonetics, has addressed the potentially problematic confluence of the

239

monolingual model and the assumption of an unchanging L1 has not been exam-

240

ined systematically, which leads to the following question: to what degree does

241

contemporary linguistic research adopting the monolingual model in fact reject where the speakers were living at the time of the study. However, given the phonetic implementation of ‘voiced’ stops in American English as voiceless unaspirated, it follows that Swedish speakers influenced by American English would produce Swedish voiced stops as voiceless unaspirated. Along the same lines, the growing population of proficient L2 speakers of English in Sweden might introduce further variation into results obtained on ‘native Swedish’.

12

242

the assumption of an unchanging L1? That is, do monolingual studies reliably

243

distinguish between monolingual and multicompetent language users? To address

244

this question, which has implications for both the interpretation and the replica-

245

tion of empirical findings on language, a methodological review was conducted of

246

recent behavioral linguistic studies focusing on monolingual populations.

247

3. Population sampling in behavioral linguistic research: A review

248

3.1. Methods

249

The basis for the methodological review was a corpus of linguistic studies

250

constructed from recent publications in high-impact journals. Given that the pri-

251

mary concern was with behavioral research meant to generalize to populations of

252

mature, functionally monolingual native speakers, the journals included were the

253

top two linguistics journals according to Google Scholar’s 2014 h5-index (‘the

254

largest number h such that h articles published in [the preceding five years] have

255

at least h citations each’) that primarily publish behavioral studies (i.e., studies

256

that require participants) directly related to language and whose focus is neither

257

on children nor on topics related to multilingualism (e.g., L2 learners, bilingual-

258

ism, language contact). These journals were the Journal of Phonetics (JPhon)

259

and Language and Cognitive Processes (LgCog; as of 2014, Language, Cognition

260

and Neuroscience). To further limit the scope, the review focused on a target time

261

period consisting of the first three years of the current decade (i.e., 2011–2013).

262

The final corpus comprised all, and only, studies that were ostensibly meant

263

to generalize to the population of adult monolingual native speakers of the sub-

264

ject language.2 All 363 articles published in JPhon and LgCog during 2011–2013 2

To clarify what is meant by ‘language’ in this context, this term is being used broadly to

13

265

were reviewed to determine whether the article met any of three exclusion criteria:

266

(1) reporting no novel adult data (e.g., child-focused or computational studies),

267

(2) focusing on multilingualism, broadly construed, and including no monolin-

268

gual control group, or (3) stating explicitly that the findings might not generalize

269

to monolinguals. Every article which did not meet an exclusion criterion was in-

270

cluded in the corpus, for a total of 286 studies (127 from JPhon, 159 from LgCog).

271

Data for each study were collected by consulting the abstract, introduction,

272

methods, and discussion sections of the paper, as well as the institutional affilia-

273

tions in the byline. To obtain a realistic picture of the potential for misinterpre-

274

tation and/or overgeneralization of findings, these data were compiled with two

275

simplifying assumptions. First, it was assumed, unless information reported on

276

the participants suggested otherwise, that participants recruited from the study lo-

277

cation had knowledge of the language(s) dominant in that region, where they were

278

presumed to have been living for a significant amount of time. For example, Span-

279

ish speakers recruited from the US were assumed to have knowledge of American

280

English unless it was specified that they were monolingual. Second, if it was not

281

stated explicitly where the study took place, it was assumed that the study location

282

corresponded to the institutional affiliation in the byline. In the case of multiple

283

affiliations, the study location was taken to be the one with the highest number of

284

speakers of the subject language according to Ethnologue (Lewis et al., 2015). refer to varieties that may be called ‘languages’ or ‘dialects’ by different researchers, meaning that studies focusing on so-called ‘dialects’ were eligible for inclusion in the corpus as long as the target population was not bidialectal speakers.

14

285

3.2. Results

286

The corpus (publicly accessible at https://osf.io/u7864/) was diverse in terms

287

of topic areas, methodologies, and subject languages. Given the target journals,

288

the corpus included research mainly in phonetics and psycholinguistics; however,

289

the psycholinguistic studies addressed questions related to virtually all areas of

290

linguistics (e.g., lexicon, phonology, syntax, semantics, pragmatics, reading, ges-

291

ture). The vast majority (94%) of studies involved laboratory-based experimen-

292

tal work, but there were also examples of archival, ethnographic, corpus-based,

293

and web-based studies. As for subject languages, the corpus was skewed toward

294

research on varieties of English (52% of studies), but a wide range of other lan-

295

guages was represented as well (e.g., Berber, Central Arrernte, Hmong, Sign Lan-

296

guage of the Netherlands).

297

Analysis of this corpus revealed recurring information gaps regarding the lan-

298

guage background of participants who were ostensibly monolingual native users

299

(speakers or signers) of a particular language. It was much more common for

300

participants’ language background to be left unclear (80% of studies) than de-

301

scribed unambiguously as monolingual (10% of studies). The many cases of am-

302

biguity stemmed from the fact that when researchers did not describe participants

303

as monolingual, they included no information regarding knowledge of other lan-

304

guages. Surprisingly, it was also not uncommon for there to be no clear descrip-

305

tion of participants as L1 users. For example, in 15% of studies, participants were

306

never described as ‘native’ or their native/first language was left unspecified. This

307

is noteworthy because nearly all of these studies involved linguistic tasks.

308

Although it was most common for studies to be ambiguous about the language

309

background of participants, a considerable number expressly examined partici-

15

310

pants who were multicompetent. Approximately 13% of studies included a native

311

group that regularly used or had significant knowledge of one or more additional

312

languages. Since none of these studies offered a reason for examining multicom-

313

petent rather than monolingual native users, the language situation in each of the

314

relevant speech communities was investigated to determine whether the use of a

315

multicompetent sample was consistent with the characteristics of the contempo-

316

rary population of language users. This investigation revealed that the multicom-

317

petence of nine user samples (Blackfoot, Catalan, Central Arrernte, Dutch, Gu-

318

jarati, Hindi, Oneida, Q’eqch’i, Trique) could be considered representative of the

319

norm in the speech community. However, these nine groups accounted for only

320

a minority of the aforementioned 13% of studies; discounting these cases still

321

left 10% of studies which examined multicompetent native users for no apparent

322

reason. These results thus suggest that about 1 in 10 studies targeting monolin-

323

guals actually examines multicompetent users, although note that by collecting

324

and reporting information about language background (see §5.3) these studies are

325

transparent about this, thus allowing the results to be properly interpreted.

326

For the majority of studies, data collection sites were appropriate given the

327

study’s aims and subject language(s); nevertheless, the practice of collecting data

328

from individuals living in a foreign language environment was found in 15% of

329

studies. These studies rarely addressed the implications of the language environ-

330

ment: in only two cases was the foreign language environment acknowledged as

331

a limitation, and in only one was a reason provided for collecting data in this

332

environment (logistical constraints). If the latter study is excluded along with

333

one other study that was presumably constrained by the availability of necessary

334

equipment (an electromagnetic articulometer), this leads to the conclusion that

16

335

about 1 in 7 studies of monolingual language performance is conducted in a non-

336

native environment (namely, the researchers’ location), meaning that the so-called

337

‘monolinguals’ are probably exposed to, and possibly proficient in, another lan-

338

guage that has potentially influenced the target language examined in the study.

339

3.3. Discussion of methodological review

340

In sum, a review of recent publications on adult monolingual language per-

341

formance revealed two problematic aspects of the literature in this area: (1) in-

342

sufficient description of study samples, and (2) disparities between study samples

343

and target populations. Across a range of topic areas and subject languages, re-

344

searchers were found to neglect describing participants’ language background—

345

sometimes even failing to identify their L1—making it difficult to tell precisely

346

what kind of individuals made up a study sample. Moreover, when information

347

about language background was provided, the given study sample often did not

348

match the target population (i.e., monolingual users in the native language envi-

349

ronment) due to multicompetence and/or residence in a foreign language environ-

350

ment. These characteristics of the sample were never required by the research

351

questions and rarely discussed in terms of their ramifications for the results.

352

Overall, these results present a troubling view of the interpretability (and repli-

353

cability) of published behavioral linguistic studies, but it should be noted that

354

there were exceptions to the patterns described above. First, some studies ex-

355

amined samples that clearly matched the populations they were meant to repre-

356

sent (e.g., French monolinguals in France: Abdelli-Beruh, 2012; Greek mono-

357

linguals in Greece and Australian English monolinguals in Australia: Antoniou

358

et al., 2012), even if these studies were in the minority. Second, although they did

359

not provide a holistic picture of language background that would make replication 17

360

straightforward, some studies did provide details beyond native language, which

361

were generally framed in terms of limitations on knowledge of other languages. A

362

few commented on aspects relevant to the materials used (e.g., no experience with

363

a crucial L2; no experience with lexical tone; no experience with vowel/consonant

364

harmony), while others highlighted general restrictions on experience with addi-

365

tional languages, particularly with the ambient language in a nonnative language

366

environment (e.g., low proficiency; late onset of learning).

367

In regard to studies of speakers in a nonnative language environment, one

368

pattern that stood out was an emphasis on their short length of residence (LoR) in

369

the L2 environment. This pattern was puzzling because different studies assumed

370

different notional thresholds (ranging from three months to two years) for the

371

maximum LoR participants could report and still be considered monolingual-like

372

in their L1. None of the thresholds, however, were justified explicitly, suggesting

373

that they were either arbitrary or conventional. If the most frequently observed

374

LoR threshold (namely, two years) is in fact convention, this raises the question

375

of whether this is the right convention for recruiting monolingual-like participants

376

in an L2 environment. Studies of L1 attrition have generally examined long-term

377

migrants, those who have been residing in an L2 environment for many years (de

378

Bot et al., 1991; de Leeuw et al., 2010; Schmid, 2013); there are no known studies

379

providing evidence for two years as a valid LoR threshold.3

380

Contrary to the view that it takes two years for an L1 speaker to show evidence

381

of L2 influence in their L1, findings on phonetic drift suggest that L2-influenced 3

Although Pavlenko and Jarvis (2002) report L2 influence in participants with an LoR of at

least three years, their results cannot be interpreted as evidence for an LoR threshold of three years because that is the shortest LoR included in that study (Pavlenko & Jarvis, 2002, pp. 193–4).

18

382

modifications to the L1 can occur within weeks of L2 exposure (Chang, 2012,

383

2013). However, because these L1 modifications were observed during a period

384

of formal L2 instruction, which may be special in terms of relative L2 activation

385

and/or L1 inhibition, it is possible that rapid L2 influence might be limited to

386

the situation of intense L2 engagement. When learners are removed from this

387

situation, such that active L2 use goes down and active L1 use goes back up, does

388

their L1 drift back to monolingual norms, or does it continue to diverge from these

389

norms due to continued ambient exposure to a now-familiar L2? This question

390

provided the motivation for a longer longitudinal study of L1 users living abroad

391

which could address the persistence of L1 drift after the end of L2 instruction.

392

4. Persistence of phonetic drift: An acoustic study

393

In light of the wide variation in assumptions regarding temporal constraints

394

on L1 restructuring, the present study investigated the time course of L1 phonetic

395

restructuring due to recent L2 experience (PHONETIC

396

The point of departure was the phonetic drift observed in Chang (2012, 2013). On

397

the one hand, because the drift in those studies was found during a period of in-

398

tensive L2 instruction immersing learners in the L2 for more than 30 hours/week,

399

it may have been due specifically to high L2 engagement associated with learn-

400

ing, which would imply that the L1 should drift back to monolingual norms once

401

active L2 learning ends; on the other hand, if the crucial ingredient for prolonging

402

drift is L2 exposure, then the discontinuation of active L2 learning should fail to

403

fully reverse drift as long as learners are living in an L2 environment.

404

405

DRIFT,

or ‘drift’ for short).

In the present study, the latter outcome was predicted on the basis of an I NCI DENTAL

P ROCESSING H YPOTHESIS (IPH), which addresses the degree to which 19

406

ambient linguistic input (i.e., input that is in the environment but not directed at the

407

listener) may be incidentally processed, rather than ignored. The IPH posits that

408

ambient linguistic input becomes increasingly difficult to ignore as one’s knowl-

409

edge of that language increases, consistent with the finding that auditory stimuli

410

are more distracting when they are informative as opposed to uninformative (Par-

411

mentier et al., 2010). The logic underlying the IPH is that, whereas ambient input

412

in an as-yet unknown (therefore, uninformative) L2 may be treated as ‘noise’,

413

thus avoiding at least deep linguistic processing, acquiring a ‘critical mass’ of

414

L2 knowledge (e.g., a sizable lexicon, phonological categories) leads to ambient

415

L2 input becoming potentially informative, such that it tends to be processed as

416

a linguistic stimulus activating linguistic representations. Crucially, this means

417

that ambient input in a known L2 is relatively likely to undergo some degree of

418

processing, even if it is not actively attended to.

419

Thus, even after the end of L2 instruction, the learners from Chang (2012,

420

2013) were predicted to continue showing drift while in the L2 environment be-

421

cause incidental processing of ambient L2 input would maintain a high activation

422

level of the L2. This prediction was tested by analyzing the L1 (English) speech

423

production of a subset of the participants in Chang (2012, 2013) one year after

424

the initial period of L2 (Korean) instruction. The dependent measures were the

425

same: VOT, onset f0 in the vowel following a stop, F1 , and F2 . Given the previ-

426

ous findings, there were four specific predictions regarding sustained drift toward

427

phonetic norms of Korean (summarized in Chang, 2012, pp. 253-4):

428

(1)

Since the initial drift in VOT of English voiceless stops had been pro-

429

nounced (on the order of 20 ms in novice learners), drift in VOT was pre-

430

dicted to persist, resulting in longer-than-baseline VOT for voiceless stops 20

431

(cf. VOT norms for Korean aspirated stops, about 28–39 ms longer than

432

for English voiceless stops).

433

(2)

Since the initial drift in onset f0 following English stops had also been

434

pronounced, drift in onset f0 was predicted to persist as well, resulting in

435

higher-than-baseline f0 for both voiced and voiceless stops (cf. f0 norms

436

following Korean fortis and aspirated stops, estimated to be at least 10–15

437

Hz higher than following English stops).

438

(3)

Since initial drift in F1 and F2 of the English vowel system had been subtle

439

(F1 ) or not significant (F2 ), drift in vowel formants was not predicted to

440

persist.

441

(4)

Since active use of the L2 might encourage the persistence of L2 influence,

442

it was predicted that the remnants of phonetic drift a year later would be

443

more obvious for L2 learners who continued to speak the L2 frequently

444

compared to those who spoke less frequently.

445

To address prediction (4) in particular, learners were analyzed in two groups dif-

446

fering in frequency of active L2 use.

447

4.1. Methods

448

4.1.1. Participants

449

A total of 36 L1 speakers of American English entered and completed the

450

initial five-week study reported in Chang (2012, 2013). All were recent college

451

graduates who had traveled to South Korea to teach English. They were invited to

452

participate in an additional study session approximately one year after their arrival

453

to Korea, and 17 elected to participate in this session. Two of these 17 participants 21

454

reported problems with hearing and/or speech in early childhood, which they per-

455

ceived as having been remedied with speech therapy; removing them from the

456

dataset did not affect the results, so they are included in all results reported below.

457

Self-identified native speakers of English, the 17 participants analyzed here

458

were raised primarily in the US and identified English as their strongest language

459

and at least one of the languages used at home. Two participants also spoke a her-

460

itage language (Mandarin in one case, Russian in the other), which they coiden-

461

tified with English as a native language. The other 15 participants identified only

462

English as a native language and reported speaking only English at home. All par-

463

ticipants had previously studied at least one foreign language (most often Spanish

464

or French) for a period of 1–13 years; however, only one reported significant com-

465

municative use of a foreign language (Japanese), which was often the language of

466

e-chat with friends. Thus, the majority (14/17) of participants were “functionally

467

monolingual” L1 English speakers (in the sense of Best & Tyler, 2007, p. 16: “not

468

actively learning or using an L2”), while the other three were bilingual L1 English

469

speakers.

470

Based on data from a detailed questionnaire about their year in Korea (pub-

471

licly accessible at https://osf.io/d5qzj/), participants were assigned to one of two

472

groups according to whether they showed low active use (LU) or high active use

473

(HU) of the L2 after the initial five-week study. This was done by ordering the

474

sample by self-reported L2 speaking frequency and splitting it evenly into two

475

groups. Nine participants (mean age 24.4 yr, SD 1.9; eight female) were thus

476

assigned to the LU group, and eight (mean age 23.6 yr, SD 0.7; seven female)

477

to the HU group. The groups did not differ significantly in terms of age [Welch-

478

corrected two-sample t(10.7) = 1.207, p = .254] although the group division put

22

479

all three bilinguals into the HU group. In addition to being closely matched on age

480

and gender, the two groups were similar with respect to several variables related

481

to use of English during the year: frequency of personal interactions with native

482

English speakers in Korea, frequency of phone/e-chat interactions with native En-

483

glish speakers in the US, and time spent away from Korea in an English-speaking

484

country, none of which differed significantly between groups [all ps > .05].

485

The principal difference between the LU and HU groups was in the nature of

486

their experience using Korean over the year. Whereas HU participants described

487

using Korean at home and/or work, LU participants described using mostly En-

488

glish both at home and at work. As a result, the LU group reported spend-

489

ing much less time speaking Korean (MLU = 2.2 hr/wk, MHU = 13.8 hr/wk;

490

t(7.5) = −5.679, p < .001) whereas they heard Korean around them much of

491

the time, just as the HU group did (MLU = 34.1 hr/wk, MHU = 49.6 hr/wk;

492

t(14.9) = −1.066, p = .303). The LU group’s limited active use of Korean

493

was further reflected in lower self-ratings of Korean proficiency across a range

494

of communicative tasks (MLU = 2.1/6 ≈ ‘poor’, MHU = 2.8/6 ≈ ‘fair’; t(8.8) =

495

−3.247, p = .010).

496

4.1.2. Learning context

497

In the initial five-week study, participants were enrolled in a Korean language

498

program at a Korean university. Prior to beginning this program, most LU partic-

499

ipants had received no significant exposure to Korean, so they were enrolled in an

500

elementary-level class; the two exceptions had taken Korean in college and were

501

enrolled in an intermediate-level class. Most HU participants were also enrolled

502

in an elementary-level class, with two enrolled in an intermediate-level class as in

503

the LU group. Despite the similarity in their enrollments, however, the HU group 23

504

was the mirror image of the LU group in terms of experience: most HU partici-

505

pants had received significant prior exposure to Korean, by virtue of having been

506

adopted from Korea (n = 3; mean age of adoption 0;11) and/or having studied

507

Korean in college (n = 4; total class contact hours ranging from 60 to 600).

508

Both elementary- and intermediate-level classes in the language program fol-

509

lowed the same intensive schedule over the six-week duration of the program. On

510

most weekdays, there were four hours of instruction, for a total of more than 80

511

class contact hours by the end of the program (roughly equivalent in content to one

512

semester of college-level Korean). Classes were conducted in Korean, and partic-

513

ipants lived on campus during the program; however, they stayed in a dormitory

514

with their fellow students, who were all native English speakers as well. Conse-

515

quently, the type of L2 learning environment provided in this program might best

516

be described as in between typical second language acquisition (in which learners

517

acquire the L2 naturalistically in an L2 environment) and typical foreign language

518

acquisition (in which learners study the L2 formally in an L1 environment).

519

Following the end of the language program, participants began working as

520

English teachers in various host locations, where most (eight LU and six HU par-

521

ticipants) reported receiving additional Korean instruction in the form of classes

522

and/or one-on-one tutoring. The amount of this instruction was similar between

523

groups (MLU = 45 hr, MHU = 47 hr), and participants reported spending little

524

time on self-regulated Korean study (MLU = 22 min/wk, MHU = 32 min/wk).

525

Crucially, additional Korean instruction tended to occur early in the year, such

526

that, on average, more than three months had elapsed between participants’ most

527

recent Korean class and the final study session (MLU = 3.7 mo, MHU = 5.1 mo).

24

528

4.1.3. Procedure

529

In the initial five-week study, participants completed two production experi-

530

ments (one in English, one in Korean) at the end of each of the first five weeks of

531

their language program, generally in a quiet room in their dormitory. Instructions

532

were provided in English, and the experiments were usually completed in one ses-

533

sion (in the order of Korean followed by English, with an intervening break). The

534

task was isolated word reading: participants were shown a target item, spelled in

535

the target language orthography, on screen and asked to say the item out loud upon

536

seeing a subsequent visual cue. This task was meant to elicit a relatively formal

537

register providing a strong test of L2 influence on L1 speech, as formal registers

538

have been shown to resist L2 influence in comparison to more casual registers

539

(Major, 1992). The experiments were administered on a Sony Vaio PCG-TR5L

540

laptop computer running DMDX (Forster, 2014). In both experiments, items were

541

randomized and presented once in each of four blocks, such that four tokens were

542

collected of each item. Recordings were made at 44.1 kHz and 16 bps using an

543

AKG C420 or C520 head-mounted condenser microphone, connected either to

544

the computer via an M-AUDIO USB preamp or to a Marantz PMD660 recorder.

545

In the additional study session that took place a year later, participants com-

546

pleted the two production experiments one more time in a quiet office in Seoul.

547

All other aspects of the procedure, materials, equipment, and recording specifica-

548

tions were the same as in the initial five-week study.

549

4.1.4. Materials

550

The speech materials for the English production experiment consisted of 24

551

monosyllabic English words: 16 critical and 8 filler items. Six critical items were

552

used to measure VOT of stops and onset f0 ; these items contained the same vowel 25

Table 1: Critical items used in the L1 (English) production experiment, by dependent measure.

Measures

Items

VOT, f0

bot, pot, dot, tot, got, cot

F1 , F2

heed, hid, hate, head, had, who’d, hood, hoed, hut, hawk, pot

553

/A/ to control for the effect of vowel environment on VOT and facilitate com-

554

parison with VOT norms based on similar contexts (Morris et al., 2008). Eleven

555

critical items (including one item used to measure VOT/f0 ) were used to measure

556

F1 and F2 of vowels; these items began with /h/ or an otherwise aspirated onset

557

to control for coarticulatory perturbations from an initial consonant and facili-

558

tate comparison with formant norms based on similar contexts (Hagiwara, 1997;

559

Hillenbrand et al., 1995; Peterson & Barney, 1952; Yang, 1996). The full set of

560

critical items, the same in every iteration of the experiment, is shown in Table 1.

561

The speech materials for the Korean production experiment consisted of 22

562

monosyllabic Korean items: 15 critical items and 7 fillers. All critical items con-

563

sisted of an open syllable comprising one consonant and one vowel. The items

564

used to measure VOT of stops and onset f0 comprised a stop (one of /p p* ph t t*

565

th k k* kh /4 ) followed by /a/, while those used to measure vowel formants com-

566

prised a vowel (one of /i 1 u E o 2 a/) preceded by /h/ (or an otherwise aspirated

567

onset). 4

The Korean stops are indicated here using conventional transcriptions for Korean laryngeal

categories. Note, however, that in Chang (2012, 2013), these stops are transcribed with the extended IPA diacritics for weak and strong articulations as, respectively, /p p ph t t th k k kh /. ^ "" ^ "" ^ ""

26

568

4.1.5. Acoustic analysis

569

The four acoustic measures were VOT in word-initial stops, onset f0 in the fol-

570

lowing vowel, and F1 and F2 at vowel midpoint. All measurements were taken in

571

Praat (Boersma & Weenink, 2016) on the waveform or a wide-band Fourier spec-

572

trogram with a Gaussian window shape (window length: 5 ms, dynamic range: 50

573

dB, pre-emphasis: 6.0 dB/oct).

574

The measures related to stops were VOT and onset f0 . VOT was calculated

575

by subtracting the time at the beginning of the release burst interval from the time

576

at voicing onset (the first point at which a voicing bar with clear glottal striations

577

appeared in the spectrogram). Onset f0 was calculated by taking the combined

578

wavelength of the first three regular glottal periods in the vowel and converting

579

to Hertz (Hz). The interval of three periods was demarcated on the waveform,

580

with an initial period being skipped if it was more than 33% longer or shorter

581

than the following period. Tokens in which the earliest interval of three regular

582

periods occurred more than five periods into the vowel were considered to have

583

an irregularly phonated vowel onset and were thus discarded.5

584

The measures related to vowel quality were F1 and F2 . Both formants were

585

measured automatically over the middle 50 ms of a vowel interval, which was de-

586

marcated manually at the first and last glottal striations showing formant structure

587

in the spectrogram. The analysis method was linear predictive coding, using the

588

Burg algorithm (Childers, 1978) in Praat. Parameters for the formant analysis (fre-

589

quency range, number of formants) were determined by visually inspecting a few

590

spectrograms from the given participant and adjusting the defaults until formant 5

A total of 1.2% of English tokens and 2.1% of Korean tokens were discarded for this reason

or because of other pronunciation anomalies such as coughing.

27

591

tracking was smooth and closely followed the formants visible in the spectrogram.

592

To further check the accuracy of the formant measurements, they were inspected

593

for outliers by vowel, potential errors were flagged, and spectrograms of all tokens

594

were inspected individually. When formant tracking was inaccurate, the analysis

595

parameters were adjusted; if this did not fix the tracking, then measurements were

596

taken manually on an average spectrum of the middle 50 ms of the vowel.6

597

Intra-rater reliability was examined via Pearson’s correlations, which indicated

598

that the measurements collected were highly reliable. Six months after the original

599

measurements were taken, approximately 20% of the analyzed tokens were ran-

600

domly selected and reanalyzed. This second round of measurements was closely

601

correlated with the first round for all measures [r = .92 to r = .98, ps < .001].

602

The average difference between paired VOT measurements was 3 ms; between

603

paired f0 measurements, 4 Hz; between paired F1 measurements, 7 Hz; and be-

604

tween paired F2 measurements, 15 Hz.

605

4.1.6. Statistical analysis

606

Prior to statistical analysis, the acoustic data were reorganized in two ways to

607

achieve a valid comparison of values across the LU and HU groups. First, stop

608

tokens were binned into three phonetic categories of stop voicing (‘prevoiced’,

609

‘short-lag’, ‘long-lag’) according to VOT boundaries estimated from the litera-

610

ture (Keating, 1984; Lisker & Abramson, 1964; Lisker et al., 1977), with the

611

most common phonetic voicing category for each stop type submitted to statis-

612

tical analysis: short-lag (VOT of 0–30 ms) for English voiced (and Korean fortis) 6

A total of 1.3% of English tokens and 0.7% of Korean tokens were discarded because of

pronunciation anomalies or speech errors.

28

613

stops, and long-lag (VOT > 30 ms) for English voiceless (and Korean aspirated)

614

stops.7 Second, frequency values (f0 , F1 , F2 ) were standardized by participant, by

615

calculating the participant’s mean for the given frequency component during the

616

initial five-week study and then expressing each of the participant’s raw values for

617

that frequency component as z-scores about the mean. This standardization al-

618

lowed for longitudinal analyses within individuals that could be compared across

619

individuals (of both genders) on the same scale.

620

The acoustic data were then modeled with mixed-effects linear regression us-

621

ing the lme() function in R (R Development Core Team, 2018). All of the

622

final models on the full dataset contained a random intercept for Participant (no

623

random slopes because these usually caused a model to fail to converge) and two

624

treatment-coded fixed effects: Time (i.e., weeks after the start of the initial Korean

625

language program: 1–5, 52; reference level = 1) and Group (LU, HU; reference

626

level = LU). In addition, models of stop-related measures (VOT, f0 ) included a

627

deviation-coded (meaning the contrast estimate is against the grand mean, rather

628

than the reference level) fixed effect for Place (of articulation: bilabial, velar, alve-

629

olar), while models of vowel-related measures (F1 , F2 ) included a deviation-coded

630

fixed effect for Vowel (/i I e E æ u U o 2 A O/), since these factors have been shown 7

Stop tokens were divided in this way primarily to exclude prevoicing (VOT < 0 ms) from the

analysis of voiced stops. Although relatively infrequent, prevoiced tokens represented a different phonetic voicing category than short-lag tokens, so to obtain a clear picture of within-group change and between-group differences in voiced stop production—one that did not simply reflect change in the frequency or robustness of closure voicing—the analysis of voiced stops was limited to tokens representing their typical short-lag realization. In the interest of consistency, the analysis of voiceless stops was also limited to tokens representing their typical long-lag realization (resulting in hardly any exclusions because nearly all tokens of voiceless stops had VOT longer than 30 ms).

29

631

to influence the given dependent variables (Ladefoged, 2005; Nearey & Rochet,

632

1994). All models further included a deviation-coded fixed effect for Block (in

633

the experiment: 1–4), representing the potential influence of fatigue or practice,

634

and all possible interactions among predictors. Block and interactions with Block

635

did not have a significant effect in any model and are thus not discussed further

636

below.

637

Due to the unbalanced distribution of talker gender and the uncertain status

638

of the /A/-/O/ merger in participants’ vowel systems, models of the frequency

639

measures underwent additional scrutiny to check that the results were robust. In

640

regard to gender, this factor was observed to have a significant effect on change

641

in frequency measures in Chang (2012, 2013); for example, females showed a

642

significant change in f0 , but males did not. However, gender could not be entered

643

into the models in the current study because there was only one male participant in

644

each group. Thus, two models of each of the frequency measures were compared:

645

one built on the full dataset, and one built on female-only data. In regard to the

646

vowel /O/, most participants (even those whose native dialect purportedly shows

647

an /A/-/O/ merger) did not show a clear merger of /O/ with /A/. Thus, two models

648

of each formant measure were compared: one including both /A/ and /O/ (i.e.,

649

assuming that participants maintained a contrast between these vowels), and one

650

including only /A/. Both cases of model comparison showed no significant effect

651

of the relevant data exclusion on the results, so the models reported below are

652

those built on the full dataset without exclusions. All data (including both the

653

acoustic data collected in the production experiments and the background data

654

coded from questionnaires) are publicly accessible at https://osf.io/u7864/.

30

110

120

LU: aspirated HU: aspirated

70

10

80

90

100

VOT (ms)

15

VOT (ms)

LU: voiceless HU: voiceless

130

LU: fortis HU: fortis

20

LU: voiced HU: voiced

1

2

3

4

5

52

1

Time (weeks in Korea)

2

3

4

5

52

Time (weeks in Korea)

(a) English voiced & Korean fortis

(b) English voiceless & Korean aspirated

Figure 1: Change in VOT of (a) English voiced and Korean fortis, and (b) English voiceless and Korean aspirated stops. The low (LU) and high active L2 use (HU) groups are shown in circles and triangles, respectively. Error bars indicate 95% confidence intervals of the mean over participants. For reference, mean VOT norms for the Korean fortis and aspirated stops (averaging over all places of articulation) are, respectively, 11–17 ms and 90–97 ms (Chang, 2012, p. 253).

655

4.2. Results

656

4.2.1. Phonetic drift in VOT

657

Consistent with the findings of Chang (2012, 2013), the VOT of English voiced

658

stops did not drift significantly in the LU or HU group (see Figure 1a). The

659

model of VOT in voiced stops is shown in the supplementary material (Table 2),

660

which also provides model summaries for all of the main models discussed below.

661

An analysis of variance (ANOVA) on the model of VOT in voiced stops (using

662

Anova() in the car package in R; Fox et al., 2018) revealed a significant effect

663

of Place [χ2 (2) = 538.789, p < .0001], which was due to bilabials having shorter-

664

than-average VOT [β = −3.944, t = −5.575, p < .0001] and velars having

31

665

longer-than-average VOT [β = 4.346, t = 5.901, p < .0001]. However, there was

666

no effect of Time [χ2 (5) = 7.987, p = .157] or Group [χ2 (1) = 0.009, p = .923].

667

Furthermore, no interactions were significant, including the Time x Group inter-

668

action [χ2 (5) = 7.498, p = .186]. As a point of comparison, Figure 1a plots

669

the VOT of the perceptually similar Korean fortis stops as well, showing that, at

670

nearly every time point, both groups produce a distinction between the English

671

and Korean stop series.

672

In contrast to the VOT of voiced stops, the VOT of English voiceless stops

673

lengthened over time. This drift in VOT was found in both groups, but persisted

674

through week 52 only in the HU group (Figure 1b). An ANOVA on the model of

675

VOT in voiceless stops revealed a significant effect of Place [χ2 (2) = 23.919, p <

676

.0001], Time [χ2 (5) = 43.794, p < .0001], and Group [χ2 (1) = 7.570, p = .006].

677

The effect of Place here was similar to the Place effect for voiced stops, while

678

the effect of Group was due to the HU group showing significantly shorter VOTs

679

overall than the LU group [β = −26.733, t = −3.815, p = .002]. The effect

680

of Time reflected a longitudinal lengthening of VOT, which for the LU group

681

was significant in week 5 [β = 8.265, t = 2.790, p = .005] but not week 52

682

[β = −1.637, t = −0.553, p = .581]. The only significant interaction was the

683

Time x Group interaction [χ2 (5) = 28.314, p < .0001]; this interaction arose

684

because VOT drifted more in the HU group than in the LU group in weeks 2, 3,

685

4, and 52 [βs > 8.999, ts > 2.087, ps < .05], although this may be due to the fact

686

that the HU group started with shorter VOT than the LU group in week 1.8 The 8

Because the LU and HU groups were based on self-reported L2 use after week 5, there was

no particular expectation regarding how the two groups would compare before week 5. However, the early divergence between groups here, particularly in week 1, merits an explanation. This may

32

687

end result was thus that only the HU group’s mean VOT remained longer in week

688

52 than in week 1. This group disparity did not appear to be due to differences in

689

L2 development: as shown in Figure 1b, the LU group showed a trajectory for the

690

similar Korean aspirated stops that resembled the HU group’s, yet only the HU

691

group showed a VOT increase from week 5 to 52 in the English voiceless stops.

692

In short, LU and HU participants were consistent in showing no drift in VOT

693

of voiced stops, but significant drift in VOT of voiceless stops during L2 instruc-

694

tion (i.e., weeks 1–5). Only HU participants, however, continued to produce

695

voiceless stop VOTs in week 52 that were significantly longer than in week 1.

696

These results thus suggest that prolongation of drift in VOT is driven not simply

697

by continued L2 exposure, but the combination of L2 exposure and active use. have to do with a global pattern evident across all of the dependent variables: in general, the HU group seems to establish more of a distance between the L1 and L2 than the LU group. Consequently, it is possible that the between-group difference in English voiceless stop VOT evident in week 1 is due at least in part to the fact that, whereas the HU group produces the English voiceless stops with shorter VOT than the Korean aspirated stops (in line with norms, given that the English stops are supposed to have shorter VOT than the Korean stops), the LU group produces them with exceedingly long VOT that appears to be “piggybacking” on the long VOTs produced for the Korean stops. On the other hand, the convergence of the groups at week 52 appears to be due to the confluence of two developments from week 5 to week 52: (1) the LU group’s decrease in VOT (for both the Korean and the English stops), and (2) the HU group’s continued increase in VOT for the English stops (despite a similar decrease in VOT for the Korean stops as seen in the LU group). Thus, it may be the case that, between week 5 and week 52, the HU group became more like the LU group in terms of conflating the English voiceless and Korean aspirated stops.

33

698

4.2.2. Phonetic drift in f0

699

In accordance with the prediction of upward drift in f0 , onset f0 follow-

700

ing English voiced stops increased from week 1 to 5 in both the LU and HU

701

groups; from week 5 to 52, however, f0 decreased in both groups, although it

702

tended to decrease less in the HU than the LU group (Figure 2a). An ANOVA

703

on the model of f0 following voiced stops showed a significant effect of Place

704

[χ2 (2) = 7.028, p = .030] and Time [χ2 (5) = 67.640, p < .0001], but no effect

705

of Group [χ2 (1) = 1.648, p = .199]. The effect of Place was unexpected and

706

not apparent in the coefficients of the main model; however, a single-predictor

707

model treating Place as a treatment-coded factor (reference level ‘alveolar’) re-

708

vealed that the source of the effect was bilabials showing a lower f0 than alveolars

709

[β = −0.130, t = −2.251, p = .025].9 The effect of Time reflected an inverse

710

U-shaped pattern of f0 drift, which for the LU group resulted in higher onset f0

711

(relative to week 1) in weeks 2–5 [βs > 0.285, ts > 2.462, ps < .05] as well as

712

week 52 [β = 0.369, t = 3.224, p = .001]. The only significant interaction was

713

the Time x Group interaction [χ2 (5) = 18.871, p = .002], which arose primarily

714

due to a tendency for the HU participants’ f0 in weeks 3–4 to evince less drift

715

relative to week 1 than LU participants’ [βs < −0.356, ts < −2.108, ps < .05].

716

However, a follow-up model built just on the HU group’s data confirmed that, as

717

in the LU group, the HU group’s f0 remained higher in week 52 than in week 9

A similar effect of place of articulation is observed in some, but not all, tones in Taiwanese,

where the clearest effect of place is velars showing the highest f0 of all (Lai et al., 2009). Lai et al. hypothesize that the higher f0 after velars may be due to larynx raising associated with tongue back raising; however, since alveolars do not involve tongue back raising, this is unlikely to cause f0 to be higher after alveolars. Thus, the cause of the Place effect observed here remains unclear.

34

1

2

3

4

5

1.5

52

0.5 0.0

1

Time (weeks in Korea)

LU: aspirated HU: aspirated

1.0

LU: voiceless HU: voiceless

-0.5

-0.5

0.0

0.5

1.0

F0 onset, standardized (z-scores)

1.5

LU: fortis HU: fortis

-1.0

F0 onset, standardized (z-scores)

LU: voiced HU: voiced

2

3

4

5

52

Time (weeks in Korea)

(a) English voiced & Korean fortis

(b) English voiceless & Korean aspirated

Figure 2: Change in f0 following (a) English voiced and Korean fortis, and (b) English voiceless and Korean aspirated stops. The low (LU) and high active L2 use (HU) groups are shown in circles and triangles, respectively. Error bars indicate 95% confidence intervals of the mean over participants. For reference, onset f0 norms for the Korean fortis and aspirated stops, due to the f0 elevation associated with laryngeally marked stop types in Korean, are expected to be higher than for the English stops.

718

1 [β = 0.275, t = 2.282, p = .023], by a margin of about 7 Hz. At every time

719

point and in both groups, the similar Korean fortis stops were produced with much

720

higher f0 than the English voiced stops, thus providing the impetus for the voiced

721

stops to drift upwards in f0 (Figure 2a).

722

As with voiced stops, onset f0 following English voiceless stops increased

723

from week 1 to 5, and then decreased from week 5 to 52, in both the LU and

724

HU groups; however, the week 5-to-52 decline was smaller in the HU than the

725

LU group (Figure 2b). An ANOVA on the model of f0 following voiceless stops

726

showed a significant effect of Time [χ2 (5) = 120.115, p < .0001], which re-

727

flected a general pattern of upward drift in weeks 2–5; for the LU group, this drift 35

728

resulted in f0 being significantly higher (compared to week 1) in all following

729

weeks including week 52 [βs > 0.289, ts > 3.120, ps < .01]. There was again an

730

effect of Place [χ2 (2) = 6.051, p = .049], similar to that found for voiced stops,

731

but no effect of Group [χ2 (1) = 1.328, p = .249]. The only significant interaction

732

was the Time x Group interaction [χ2 (5) = 27.211, p < .0001], which, as for

733

voiced stops, was due to HU participants’ f0 in weeks 3–4 drifting less relative

734

to week 1 than LU participants’ [βs < −0.292, ts < −2.168, ps < .05]. Cru-

735

cially, however, a follow-up model built just on the HU group’s data confirmed

736

that the HU group’s f0 remained significantly higher in week 52 than in week 1

737

[β = 0.406, t = 4.395, p < .0001], by a margin of about 9 Hz. At every time

738

point and in both groups, especially the HU group, the similar Korean aspirated

739

stops were produced with higher f0 than the English voiceless stops (Figure 2b).

740

Thus, the LU and HU groups both showed drift in onset f0 during L2 in-

741

struction, as well as a decline in this drift after L2 instruction. In both groups,

742

however, the decline was incomplete, with f0 remaining elevated above week 1

743

levels in week 52. Given that sustained drift in VOT was found only in the HU

744

group, these results therefore suggest that drift in features related to f0 level may

745

be more persistent than drift in VOT. In particular, prolongation of drift in onset

746

f0 does not appear to require extensive active L2 use as in the HU group.

747

4.2.3. Phonetic drift in F1 and F2

748

The evolution of the English vowel space over time is shown in Figures 3–4

749

(omitting weeks 2–4 for clarity) for the LU and HU groups, respectively. These

750

figures show that although longitudinal shifts in individual vowels were generally

751

subtle, there was systematic change at the level of the system with respect to both

752

F1 and F2 ; however, this change was more evident in the LU than the HU group. 36

753

An ANOVA on the model of F1 revealed the expected effect of Vowel [χ2 (10) =

754

69954.072, p < .0001], as well as an effect of Time [χ2 (5) = 18.154, p = .003];

755

however, there was no effect of Group [χ2 (1) = 1.945, p = .163]. The effect of

756

Time reflected an overall pattern of F1 decrease from week 1 to week 5, followed

757

by F1 increase from week 5 to 52, which tracked quite closely the pattern in

758

Korean (Figure 5a). For the LU group, this drift pattern ended in mean F1 being

759

significantly higher in week 52 than in week 1 [β = 0.072, t = 3.980, p < .001].

760

Note, however, that the higher F1 in week 52 does not necessarily represent an

761

overcompensation for the downward drift in F1 since the true baseline corresponds

762

to week 0, which was not observed (i.e., it is possible that F1 in week 52, although

763

higher than F1 in week 1, does not differ from the baseline F1 in week 0).

764

In addition to the main effects of Vowel and Time, there were three signifi-

765

cant interactions: Vowel x Group [χ2 (10) = 26.091, p = .004], Time x Group

766

[χ2 (5) = 21.469, p < .001], and Vowel x Time x Group [χ2 (50) = 89.987, p <

767

.001]. The Vowel x Group interaction was primarily due to the vowel /A/, pro-

768

duced with higher F1 (relative to the center of the vowel space) in the HU than

769

the LU group [β = 0.118, t = 2.074, p = .038]. The Time x Group interac-

770

tion reflected the relatively flat pattern of F1 drift in the HU group—in particu-

771

lar, the smaller F1 increase between weeks 5 and 52 compared to the LU group

772

[β = −0.062, t = −2.430, p = .015]. To explore this interaction further, an ad-

773

ditional model (with the same structure as the main model but no Group factor)

774

was built on just the HU group’s data. An ANOVA on this model indicated that

775

the Time x Group interaction arose because, unlike the LU group, the HU group

776

did not show a significant effect of Time on F1 [χ2 (5) = 4.684, p = .456]. Fi-

777

nally, the Vowel x Time x Group interaction was due to several vowels (/i æ o A/)

37

o ɪ ʊ

0.5

0.0

-0.5 -1.0 -1.5

u e

ɛ

1.0

ʌ

1.5

ɔ æ

Week 1 Week 5 Week 52

2.0

F1, standardized (z-scores)

i

2.0

1.5

1.0

0.5

ɑ

0.0

-0.5 -1.0 -1.5

F2, standardized (z-scores) Figure 3: F1 by F2 of English vowels over time for the low active L2 use (LU) group. Week 1 means are shown with squares and solid gray lines; Week 5 means, with circles and dotted gray lines; and Week 52 means, with triangles and solid black lines. Error bars indicate ±1 mean standard error.

38

o

ɪ

ʊ

0.0

-0.5 -1.0 -1.5

u

e

ʌ

1.5

1.0

0.5

ɛ

2.0

1.5

1.0

ɔ

æ

Week 1 Week 5 Week 52

2.0

F1, standardized (z-scores)

i

0.5

ɑ

0.0

-0.5 -1.0 -1.5

F2, standardized (z-scores) Figure 4: F1 by F2 of English vowels over time for the high active L2 use (HU) group. Week 1 means are shown with squares and solid gray lines; Week 5 means, with circles and dotted gray lines; and Week 52 means, with triangles and solid black lines. Error bars indicate ±1 mean standard error.

39

0.2

LU: Korean HU: Korean

-0.2

0.0

LU: English HU: English

-0.6

-0.4

-0.4

-0.2

0.0

F2, standardized (z-scores)

LU: Korean HU: Korean

-0.6

F1, standardized (z-scores)

0.2

LU: English HU: English

1

2

3

4

5

1

52

2

3

4

5

52

Time (weeks in Korea)

Time (weeks in Korea)

(a) F1 (standardized)

(b) F2 (standardized)

Figure 5: Mean (a) F1 and (b) F2 of the English and Korean vowel systems over time, by group. The low (LU) and high active L2 use (HU) groups are shown in circles and triangles, respectively. Error bars indicate 95% confidence intervals of the mean over participants. For reference, mean F1 and F2 norms for Korean vowels (averaging over the vowel inventory) are consistently lower than for English vowels due to the smaller number of open and front vowels in Korean (Chang, 2012, p. 254).

778

patterning differently in week 2 in the HU group compared to the LU group.

779

As for F2 , here, too, there was significant drift, but again only in the LU group;

780

furthermore, the pattern of drift was increasing (as opposed to mostly decreasing

781

for F1 ). An ANOVA on the model of F2 again revealed a significant effect of

782

Vowel [χ2 (10) = 98327.083, p < .0001] and Time [χ2 (5) = 20.437, p = .001],

783

but no effect of Group [χ2 (1) = 1.667, p = .197]. The effect of Time was due

784

to a longitudinal increase in F2 for LU learners, which resembled their trajectory

785

in Korean (Figure 5b) and resulted in significantly higher F2 (relative to week

786

1) in all following weeks [βs > 0.032, ts > 2.099, ps < .05]. By contrast, an

40

787

additional model built on just the HU group’s data (with the same structure as the

788

main model but no Group factor) showed no significant effect of Time [χ2 (5) =

789

8.300, p = .141], reflecting the fact that F2 in the HU group did not significantly

790

differ in weeks 2–52 from week 1 levels [βs < 0.029, ts < 1.893, ps > .05]. The

791

only significant interaction in the main model was the Vowel x Group interaction

792

[χ2 (10) = 36.204, p < .0001], due to /U/ being produced with lower F2 (relative

793

to the center of the vowel space) in the HU than the LU group [β = −0.101, t =

794

−2.082, p = .037].

795

In sum, the LU group, but not the HU group, showed drift in F1 and F2 . Drift

796

in F1 occurred via a decrease between weeks 1 and 5, followed by an increase be-

797

tween weeks 5 and 52, whereas drift in F2 occurred overall via an increase from

798

week 2 onwards. Although not all vowels moved in a manner consistent with the

799

overall patterns, the observed effects were not isolated to just a few vowels, as

800

reflected in the non-significance of the Vowel x Time interaction in all models.

801

Notably, these results, vis-a-vis the VOT results, show the reverse group disparity,

802

suggesting that prolongation of drift in vowel formants is not dependent on fre-

803

quent active L2 use and, moreover, that frequent active L2 use might actually play

804

a role in increasing the stability of the L1 vowel space in the face of ambient L2

805

exposure.

806

5. General discussion

807

5.1. Synthesis of the findings

808

Results of the longitudinal study were consistent with the I NCIDENTAL P RO H YPOTHESIS (IPH) that ambient input in a familiar L2 would tend to

809

CESSING

810

be processed, thereby promoting high L2 activation: phonetic drift of the L1 dur41

811

ing L2 instruction persisted post-instruction within the L2 environment. However,

812

the specific predictions based on findings in Chang (2012, 2013) were only par-

813

tially supported. As expected, voiceless stops drifted in VOT and both voiced and

814

voiceless stops drifted in onset f0 during initial L2 instruction; while the VOT

815

drift persisted only among frequent L2 speakers (cf. predictions (1) and (4)), the

816

f0 drift persisted among both frequent and less frequent L2 speakers (cf. predic-

817

tion (2)). Unexpectedly, vowels underwent drift in both F1 and F2 among less

818

frequent L2 speakers, and this drift persisted after L2 instruction (cf. prediction

819

(3)). Also unexpected was the fact that extensive active L2 use was associated

820

with sustained drift in VOT of consonants, but with resistance to drift in F1 and

821

F2 of vowels (cf. prediction (4)). Together, these results provide evidence that, in

822

one or more ways, the L1 production of L2 learners tends to diverge from mono-

823

lingual L1 norms during L2 instruction, and then tends to stay that way in an L2

824

environment, even when learners do not continue to speak the L2 very frequently.

825

In evaluating the current findings, it is worth noting that these conclusions are

826

on the conservative side, since weeks 2–52 were compared to week 1 (i.e., L1

827

production after one week of L2 learning), not to week 0 (i.e., true baseline L1

828

production). For the one case of apparent reversal of drift by week 52 (i.e., VOT

829

in the LU group), it is therefore possible that the data from week 52 represent

830

only partial reversal (returning to a week 1 level already significantly different

831

from baseline) rather than full reversal (returning all the way to baseline). The

832

fact that true baseline L1 production was not able to be observed is a limitation

833

of this study that does not allow for the conclusion of full reversal of phonetic

834

drift. Crucially, however, the data observed from week 1 onwards support the

835

conclusion that even L2 learners who report limited active use of the L2 tend to

42

836

differ from L1 monolinguals while in an L2 environment, which argues against the

837

methodological conflation of monolingual and multicompetent ‘native speakers’.

838

An additional limitation of this study is the fact that the LU and HU groups

839

differed in at least two ways besides active L2 use after week 5. Recall from

840

§4.1.1 that, although the groups were matched along a variety of demographic

841

and experiential dimensions, they differed in terms of prior exposure to Korean

842

and bilingualism. The fact that the HU group contained more individuals with

843

prior exposure to Korean, including Korean Americans who were adopted from

844

Korea at an early age, as well as some bilinguals (in contrast to the LU group) is

845

relevant given the evidence that international adoptees retain knowledge of their

846

birth language, which may confer an advantage in (re)learning (Bowers et al.,

847

2009; Oh et al., 2010; Choi et al., 2017), and that bilinguals also may have an

848

advantage in phonetic learning (Antoniou et al., 2015). Post-hoc analyses did not

849

actually reveal systematic differences between adoptee or bilingual members of

850

the HU group and the other members, suggesting that they were not solely re-

851

sponsible for the patterning of the HU group. Nevertheless, the group differences

852

observed in §4—in particular, in week 52—should be taken with the proverbial

853

grain of salt, as it cannot be guaranteed that they were solely due to the basis of

854

the group division (i.e., self-reported active L2 use after week 5).

855

Although the current findings are consistent with those reported in Chang

856

(2012, 2013) for drift in VOT and f0 , they differ with respect to drift in vowel

857

formants. In particular, the (mostly female) LU group in this study showed drift

858

in both F1 and F2 , whereas the female learners in Chang (2012, 2013) showed

859

drift in F1 , but not in F2 . In fact, the upward drift in F2 exhibited by the LU group

860

in this study resembles the upward drift in F2 exhibited by the male learners in

43

861

Chang (2012, 2013); however, whereas the drift exhibited by those male learners

862

can be interpreted as convergence toward the L2 at an acoustic level (in approxi-

863

mation to an L2 model based on female instructors), the drift exhibited by the LU

864

group in the current study cannot be interpreted as convergence, since the increase

865

in F2 had the effect of taking the L1 vowel system further away from the lower F2

866

level of the target L2 vowel system (Figure 5b; see also Chang, 2012, p. 254).

867

Why, then, did the learners in this study show a pattern of drift in F2 that was

868

effectively dissimilatory vis-a-vis the L2? Further research is needed to answer

869

this question, but one factor that may be playing a role is the crosslinguistic con-

870

vergence resulting from the drift in F1 . In other words, perhaps F2 drifted in the

871

observed manner in response to the drift that occurred in F1 , diverging from the

872

L2 in order to keep a certain amount of distance between the L1 and L2 vowel sys-

873

tems; such a concern for maintaining crosslinguistic contrast would be consistent

874

with the SLM principle of a shared phonetic space for L1 and L2 sounds (Flege,

875

1995; see also Lang & Davidson, in press). Regardless of how this drift pat-

876

tern is interpreted, however, the basic finding of F2 drift among less frequent L2

877

speakers—sustained well after L2 instruction—lends further support to the claim

878

that L1 users with L2 experience are different users of the L1 than monolinguals.

879

To my knowledge, this is the first study to track L1 phonetic developments in

880

L2 learners in relation to temporal separation from L2 instruction and frequency

881

of continued active L2 use. As such, apart from the IPH, there is no established

882

theory that applies directly to all of the different acoustic properties examined

883

here. The main contribution of the current set of results, therefore, is not in test-

884

ing the predictions of an existing theory, but in paving the way for further work

885

in this area, which may lead ultimately to an elaborated theory of phonetic drift

44

886

as a function of variables such as acoustic dimension, amount/type/timing of L2

887

experience, and cognitive and affective factors. At this point in time, one can ap-

888

peal to certain explanations for some of the variation in drift seen in this study; for

889

example, a control mechanism for f0 that is distinct from segment-level control

890

mechanisms and shared across languages might be behind the VOT-f0 disparity

891

(cf. Chang, 2010). However, given the modest sample size as well as participants’

892

unique backgrounds as language teachers, it should be borne in mind that more

893

research is needed to understand the extent to which the current results will gen-

894

eralize to other L2 users.10 10

An anonymous reviewer wondered, for instance, about participants’ exposure to Korean-

accented English as language teachers in L2 English classrooms; English teachers in Korea may interact with L2 English interlocutors outside of the classroom as well. Could contact with Koreanaccented interlocutors have caused the observed drift of participants’ English, which usually converged toward acoustic properties of Korean? While the potential role of such contact cannot be completely excluded, research suggests, on the contrary, that native interlocutors tend to diverge from nonnative interlocutors, at least those who are strongly accented and who are not particularly close to them socially (Kim, 2009; Kim et al., 2011). More generally, there are also characteristics of language teachers that disfavor their assimilating an L2 accent resulting in deviation from L1 norms. L1 users who gravitate toward language teaching may do so because they have an “instructional orientation” toward the L1, which may be related to, and/or enhanced by, high metalinguistic awareness and explicit knowledge of rules, norms, and standards. Additionally, teaching one’s L1 involves the unique production experience of repeatedly articulating the L1 in a clear, careful, standard manner. Consequently, although the need to examine the generalizability of these results remains, I regard it as unlikely for the current participants to have assimilated the observed drift directly from Korean-accented English speakers.

45

895

5.2. Language change over the lifespan

896

In the context of a growing body of research in language variation and change

897

showing L1 developments occurring well after childhood (Harrington et al., 2000;

898

Sankoff & Blondeau, 2007; Wagner & Sankoff, 2011; Wagner, 2012; Rickford &

899

Price, 2013), the contribution of the present study is in highlighting the role of

900

L2 experience in lifespan change. The data in §4.2 suggest that L2 contact does

901

not need to build to a high level of proficiency or involve extensive active use in

902

order to have a detectable effect on the L1. On the contrary, once L2 learners

903

have acquired a modicum of L2 knowledge, continued ambient exposure to the

904

L2 exerts a significant influence on some (but not all) aspects of L1 production,

905

and continued active L2 use appears to enhance this effect in certain cases. These

906

findings support the dynamic view of L1 knowledge that follows from the multi-

907

competence framework (Cook, 1992, 2003) and the dynamic systems approach to

908

lifespan development (de Bot, 2007), arguing in favor of giving thoughtful consid-

909

eration to language background, including recent L2 experience, as a component

910

of behavioral linguistic research. To provide a more comprehensive view of the

911

persistence of L1 drift, future studies could track developments in L1 production

912

after L2 learners have returned to an L1 environment (cf. Sancier & Fowler, 1997).

913

Of course, lifespan change is not necessarily limited to language knowledge,

914

but may extend to socio-affective dimensions such as identity and group affili-

915

ation. Thus, it is worth noting that, apart from the cognitive pressure favoring

916

L2-influenced change in the L1, such change may also be influenced by the social

917

signaling potential of manifesting L2-like features (see, e.g., Sharma & Sankaran,

918

2011; Alam & Stuart-Smith, 2011, 2014). Given that ‘Korean English’ did not

919

have the status of a regional ethnic English within Korea at the time of the cur-

46

920

rent study and, moreover, participants were often the only L1 English speaker in

921

their locality, it is not clear that participants’ manifesting Korean-like features in

922

their English would have served a coherent socio-indexical function in this case.

923

Nevertheless, one can imagine how the right conditions could arise such that L1

924

English teachers in Korea begin to identify as part of a unique, ‘L1 English expat

925

in Korea’ community, resulting in Korean-colored English acquiring social mean-

926

ing (e.g., ‘localness’) that encourages the increasing use of Korean-like features as

927

this community sets itself apart from more short-term English-speaking visitors.

928

In addition to the role of sociolinguistic factors, another direction for future

929

research on L2-influenced L1 change is its time course. Although some results

930

in §4.2 indicate sustained drift, a portion of the results also indicate that L2-

931

influenced changes in the L1 can be short-lived, dissipating in the absence of

932

frequent active L2 use. This type of finding is consistent with some of the results

933

reviewed in §2.1 as well as other results, such as the lack of effect of variation

934

in L2 Spanish use on L1 Quichua accent (Guion et al., 2000; cf. Yeni-Komshian

935

et al., 2000 and de Leeuw et al., 2010). The existence of weak or null effects of

936

L2 experience thus brings us back to the question of constraints on L2 influence.

937

Given that previously proposed constraints do not seem to hold consistently (see

938

§2.1), much more research will be required to understand when L2 effects persist,

939

where they come from (e.g., learning, use, and/or exposure), and how they differ

940

according to the type of L1 structure or property at issue.

941

In connection with the latter question, a central concern for future studies of

942

L2-influenced L1 change will be accounting for when L2 influence manifests as an

943

ostensibly ‘negative’ effect (i.e., resulting in divergence from monolingual norms)

944

and when it does not. Complementing the abundance of ‘negative’ L2 effects sum-

47

945

marized in §2.1, ‘positive’ effects resulting in some kind of advantage over mono-

946

linguals are also reported in the literature. In addition to the domain-general and

947

metalinguistic benefits of bilingualism, L2 learning has been linked to production

948

of greater L1 complexity (Kecskes, 1998) as well as less apparent attrition of the

949

L1. For example, L1 Russian immigrants in Israel proficient in L2 Hebrew per-

950

form more similarly to Russian monolinguals in judging the correctness of com-

951

plex grammatical constructions in Russian than do Russian immigrants who do

952

not know Hebrew (Laufer & Baladzhaeva, 2015). Similarly, in the present study,

953

frequent L2 speakers had an advantage over less frequent L2 speakers in L1 vowel

954

stability. Therefore, it would be a gross oversimplification to say that L2 learning

955

necessarily ‘interferes’ with the L1, because it is clear that the consequences of

956

multicompetence show a variability that is not yet fully understood.

957

5.3. Best practice in treatment of language background

958

The methodological review in §3 revealed a tendency for behavioral studies

959

in linguistics to contain vague definitions of target populations and/or mismatches

960

between target populations and participant samples, reflecting an overreliance on

961

nativeness to define language background despite the fact that the term ‘native’ is

962

not a precise descriptor. Because language histories can be complex (in particular,

963

multilingual), omission of this information from a study report implies that it is

964

irrelevant to the aims or results of the study. In light of the findings in §2, however,

965

it is not clear that any type of linguistic behavior can be safely assumed to remain

966

unaffected by multilingualism. On the contrary, an abundance of evidence—not

967

only the phonetic data in the present study but also the extensive findings reviewed

968

in §2.1—suggests that L1 knowledge remains, to some degree, plastic across the

969

lifespan and, in particular, responsive to changes in the L1 user’s circumstances. 48

970

This view of the adult L1 system as dynamic, as opposed to immutable, high-

971

lights both the empirical inadequacy of the category ‘native’ as well as the need

972

to move toward more informative descriptions of language users. Because re-

973

sults found with one kind of native speaker may not generalize to a different kind

974

(e.g., one with a different language background or belonging to a different speech

975

community), specifying a target population only in terms of the broad category

976

of ‘native’ is likely to complicate attempts at replication as long as the L1 is as-

977

sumed to be unchanging. For example, if a result reportedly obtained with ‘native’

978

speakers of Canadian English fails to be replicated with Canadian English mono-

979

linguals, it is difficult to know how to interpret this: is it actually evidence against

980

the original result or just the product of a sampling difference with respect to the

981

original study (which, given the trends discussed in §3.2, may have been based on

982

English-French bilinguals or speakers residing outside Canada)?

983

Importantly, it should be noted that the kind of considered treatment of lan-

984

guage background that will aid future research, including replication, does not

985

have to be complicated and is already supported by a number of published re-

986

sources. There are, for example, several well-described instruments for collect-

987

ing data about language history and background, such as the Language Experi-

988

ence and Proficiency Questionnaire (Marian et al., 2007), the Bilingual Language

989

Profile (Birdsong et al., 2012), the Bilingual Language Experience Calculator

990

(Unsworth, 2013), the Language History Questionnaire (Li et al., 2014), and the

991

custom-designed questionnaire for heritage Korean speakers used by Ahn et al.

992

(2017). Such questionnaires do not take very long to administer and provide a

993

wealth of data about participants’ language backgrounds, which are useful for de-

994

termining whether study participants are representative of the target population.

49

995

However, only a fraction of this information (in particular, language proficiencies

996

and acquisition profile) typically needs to be presented for the reader to understand

997

the nature of the participant sample. That is, showing that language background

998

has been controlled for, especially when the target population is monolinguals,

999

does not require an undue amount of time or journal space.

1000

The imperative to appropriately control and describe language background

1001

invites the question of what specific variables related to language background are

1002

crucial to consider in linguistic research. In short, the answer to this question will

1003

depend on the nature of the research question(s), but it is worth bearing in mind

1004

that there are two reasons why a researcher might consider collecting and then

1005

reporting certain data about language background. The first reason is to ensure

1006

that the results obtained in the study can in fact address the research question

1007

posed. The second is to aid replication: for other researchers to run the same study

1008

(targeting the same population), they need to know enough about the participant

1009

sample from the original study to be able to put together a similar sample.

1010

Thus, apart from general information about participants’ multicompetence

1011

(i.e., knowledge and acquisition of languages beyond the target language), the

1012

aspects of language background which are relevant to focus on will differ across

1013

studies and must be identified by thinking about whether/how the L1 phenomenon

1014

under investigation might be affected by L2 experience. In the easy (and probably

1015

uncommon) case, there is already strong evidence that the given phenomenon is

1016

not affected by L2 experience. However, in the absence of such evidence, there

1017

is a need to understand both the nature of participants’ L2 experience, as well as

1018

the manner in which the target phenomenon might be affected by this L2 expe-

1019

rience (e.g., whether the potential effect goes in the direction of or against the

50

1020

hypothesis), which underscores the importance of research in L2 acquisition and

1021

L1 attrition in furthering the theory of crosslinguistic congruence (i.e., the points

1022

of overlap or similarity that may exist between different languages), convergence,

1023

and divergence, across multiple levels of language.

1024

6. Conclusion

1025

Although the monolingual model of the language user is prevalent in linguistic

1026

research, research on L2 learning and L1 attrition, including the study in §4, sug-

1027

gests that the monolingual model, in conjunction with assuming an unchanging

1028

L1, may lead to an inaccurate picture of the target speech community. Actual lan-

1029

guage users are often not monolingual, and when they are competent in additional

1030

languages, their multicompetence cannot be ignored because multicompetence is

1031

transformative, not merely additive; that is, previously acquired linguistic systems

1032

are, to some degree, plastic, rather than fixed. This view is at odds with current

1033

methodological practices in linguistic research related to participant sampling and

1034

description of monolinguals, which tend to be underinformed regarding the lan-

1035

guage background of so-called ‘native’ speakers.

1036

Thus, the present study serves as a call to the field to address the variable

1037

of language background in sufficient detail to allow behavioral research findings

1038

focused on monolinguals to be interpreted and replicated transparently. As shown

1039

in §4, L2 learning exerts a rapid and, in some cases, persistent effect on the L1

1040

even when the L2 is not spoken very frequently; consequently, this is an issue

1041

about L2 contact in general, not about high levels of L2 proficiency or frequent

1042

active L2 use in particular. For research on monolingual speakers, the way forward

1043

is to give language background the same kind of considered treatment that one 51

1044

sees only occasionally in research on monolinguals (e.g., studies cited in §3.3)

1045

but regularly in the research on bi-/multilinguals.

1046

In closing, it cannot be overemphasized that the recommendation for updating

1047

methodological practices in research on monolinguals should not be construed as

1048

a recommendation for conducting research according to the monolingual model.

1049

For a language typically spoken by multilinguals, there may be fine research ques-

1050

tions that directly engage only one of these speakers’ languages, and controlling

1051

for language background appropriately provides a reasonable way of addressing

1052

such questions; however, there are also many interesting questions to be asked

1053

about these speakers’ multingualism. Thus, whereas one approach to examining

1054

a language in this type of multilingual context is to control for language back-

1055

ground, another approach, poised to provide broader insight into such multilingual

1056

speakers, is to examine these multilinguals as multilinguals. In fact, this type of

1057

diasporic or expatriate community is not only worthy of holistic investigation, but

1058

uniquely positioned to improve our understanding of crosslinguistic interaction,

1059

language change, and language stability in a mobile, multilingual world.

1060

Acknowledgments

1061

The author gratefully acknowledges financial support from the National Sci-

1062

ence Foundation (BCS-0922652) and the Center for Korean Studies and Depart-

1063

ment of Linguistics at UC Berkeley; research assistance from Daiana Chang and

1064

Kevin Sitek; and logistical assistance from the Fulbright Korean-American Edu-

1065

cational Commission. The research reported here benefited from the feedback of

1066

many individuals, including Ocke-Schwen Bohn, Ann Bradlow, Chiara Celata,

1067

Taehong Cho, Lisa Davidson, Esther de Leeuw, Susanne Gahl, Carla Hudson 52

1068

Kam, Sharon Inkelas, Keith Johnson, John Ohala, Kathryn Pruitt, three anony-

1069

mous reviewers, and audiences at UC Berkeley, Yonsei University, NYU, the

1070

CUNY Graduate Center, Rice University, the 11th International Symposium on

1071

Bilingualism (ISB 11), and meetings of the Linguistic Society of America and the

1072

Acoustical Society of America. Any errors, however, are those of the author.

53

1073

References

1074

Abdelli-Beruh, N. B. (2012). Voicing assimilation of French /t/. Journal of Pho-

1075

netics, 40(3), 521–534.

1076

Ahn, S., Chang, C. B., DeKeyser, R., & Lee-Ellis, S. (2017). Age effects in first

1077

language attrition: Speech perception by Korean-English bilinguals. Language

1078

Learning, 67(3), 694–733.

1079

Alam, F. & Stuart-Smith, J. (2011). Identity and ethnicity in /t/ in Glasgow-

1080

Pakistani high-school girls. In Lee, W.-S. & Zee, E. (Eds.), Proceedings of the

1081

17th International Congress of Phonetic Sciences (pp. 216–219). Hong Kong:

1082

City University of Hong Kong.

1083

Alam, F. & Stuart-Smith, J. (2014). Identity, ethnicity and fine phonetic detail: An

1084

acoustic phonetic analysis of syllable-initial /t/ in Glaswegian girls of Pakistani

1085

heritage. In Hundt, M. & Sharma, D. (Eds.), English in the Indian Diaspora

1086

(pp. 29–53). Amsterdam, The Netherlands: John Benjamins Publishing.

1087

Antoniou, M., Liang, E., Ettlinger, M., & Wong, P. C. M. (2015). The bilingual

1088

advantage in phonetic learning. Bilingualism: Language and Cognition, 18(4),

1089

683–695.

1090

Antoniou, M., Tyler, M. D., & Best, C. T. (2012). Two ways to listen: Do L2-

1091

dominant bilinguals perceive stop voicing according to language mode? Jour-

1092

nal of Phonetics, 40(4), 582–594.

1093

Bang, H.-Y., Sonderegger, M., Kang, Y., Clayards, M., & Yoon, T.-J. (2018). The

1094

emergence, progress, and impact of sound change in progress in Seoul Korean:

54

1095

Implications for mechanisms of tonogenesis. Journal of Phonetics, 66, 120–

1096

144.

1097

Bassetti, B. & Cook, V. J. (2011). Relating language and cognition: The second

1098

language user. In V. J. Cook & B. Bassetti (Eds.), Language and Bilingual

1099

Cognition (pp. 143–190). Hove, UK: Psychology Press.

1100

Beinhoff, B. (2008). Looking for the “real” native speaker: The perception of

1101

native and non-native English accents by non-native speakers of English. In

1102

E. Waniek-Klimczak (Ed.), Issues of Accents in English (pp. 120–139). Cam-

1103

bridge, UK: Cambridge Scholars Publishing.

1104

Bergmann, C., Nota, A., Sprenger, S. A., & Schmid, M. S. (2016). L2 immer-

1105

sion causes non-native-like L1 pronunciation in German attriters. Journal of

1106

Phonetics, 58, 71–86.

1107

Best, C. T. & Tyler, M. D. (2007). Nonnative and second-language speech per-

1108

ception: Commonalities and complementarities. In O.-S. Bohn & M. J. Munro

1109

(Eds.), Language Experience in Second Language Speech Learning: In Honor

1110

of James Emil Flege (pp. 13–34). Amsterdam, The Netherlands: John Ben-

1111

jamins Publishing.

1112

1113

Bialystok, E. & Craik, F. I. M. (2010). Cognitive and linguistic processing in the bilingual mind. Current Directions in Psychological Science, 19(1), 19–23.

1114

Birdsong, D., Gertken, L. M., & Amengual, M. (2012). Bilingual language pro-

1115

file: An easy-to-use instrument to assess bilingualism. Center for Open Ed-

1116

ucational Resources and Language Learning, University of Texas at Austin.

1117

https://sites.la.utexas.edu/bilingual/. 55

1118

1119

Boersma, P. & Weenink, D. (2016). Praat: Doing phonetics by computer. Version 6.0.19. http://www.praat.org.

1120

Bournot-Trites, M. & Tellowitz, U. (2002). Report of Current Research on the Ef-

1121

fects of Second Language Learning on First Language Literacy Skills. Halifax,

1122

Canada: The Printing House.

1123

1124

Bowers, J. S., Mattys, S. L., & Gage, S. H. (2009). Preserved implicit knowledge of a forgotten childhood language. Psychological Science, 20(9), 1064–1069.

1125

Brown, A. (2008). Gesture viewpoint in Japanese and English: Crosslinguistic

1126

interactions between two languages in one speaker. Gesture, 8(2), 256–276.

1127

Brown, A. & Gullberg, M. (2011). Bidirectional cross-linguistic influence in event

1128

conceptualization? Expressions of Path among Japanese learners of English.

1129

Bilingualism: Language and Cognition, 14(1), 79–94.

1130

Brown, A. & Gullberg, M. (2012). Multicompetence and native speaker variation

1131

in clausal packaging in Japanese. Second Language Research, 28(4), 415–442.

1132

Caramazza, A. & Yeni-Komshian, G. H. (1974). Voice onset time in two French

1133

dialects. Journal of Phonetics, 2(3), 239–245.

1134

Celata, C. (in press). Phonological L1 attrition. In M. S. Schmid & B. Köpke

1135

(Eds.), The Oxford Handbook of Language Attrition. Oxford, UK: Oxford Uni-

1136

versity Press.

1137

1138

Chang, C. B. (2010). First Language Phonetic Drift During Second Language Acquisition. PhD thesis, University of California, Berkeley, Berkeley, CA.

56

1139

1140

1141

1142

Chang, C. B. (2012). Rapid and multifaceted effects of second-language learning on first-language speech production. Journal of Phonetics, 40(2), 249–268. Chang, C. B. (2013). A novelty effect in phonetic drift of the native language. Journal of Phonetics, 41(6), 491–504.

1143

Chang, C. B. (in press). Phonetic drift. In M. S. Schmid & B. Köpke (Eds.), The

1144

Oxford Handbook of Language Attrition. Oxford, UK: Oxford University Press.

1145

Childers, D. G. (1978). Modern Spectrum Analysis. New York, NY: John Wiley

1146

& Sons.

1147

Cho, M.-H. & Lee, S. (2016). The impact of different L1 and L2 learning expe-

1148

rience in the acquisition of L1 phonological processes. Language Sciences, 56,

1149

30–44.

1150

Choi, J., Broersma, M., & Cutler, A. (2017). Early phonology revealed by interna-

1151

tional adoptees’ birth language retention. Proceedings of the National Academy

1152

of Sciences of the United States of America, 114(28), 7307–7312.

1153

1154

1155

1156

1157

1158

1159

1160

Chomsky, N. (1957). Syntactic Structures. The Hague, The Netherlands: Mouton de Gruyter. Chomsky, N. (1986). Knowledge of Language: Its Nature, Origin, and Use. New York, NY: Praeger Publishers. Cook, V. J. (1991). The poverty-of-the-stimulus argument and multicompetence. Second Language Research, 7(2), 103–117. Cook, V. J. (1992). Evidence for multicompetence. Language Learning, 42(4), 557–591. 57

1161

Cook, V. [J.] (1997). The consequences of bilingualism for cognitive process-

1162

ing. In A. M. B. de Groot & J. F. Kroll (Eds.), Tutorials in Bilingualism:

1163

Psycholinguistic Perspectives (pp. 279–300). Mahwah, NJ: Lawrence Erlbaum

1164

Associates.

1165

Cook, V. [J.] (2003). The changing L1 in the L2 user’s mind. In V. Cook (Ed.),

1166

Effects of the Second Language on the First (pp. 1–18). Clevedon, UK: Multi-

1167

lingual Matters.

1168

1169

Davies, A. (2003). The Native Speaker: Myth and Reality. Clevedon, UK: Multilingual Matters.

1170

de Bot, K. (2007). Dynamic systems theory, lifespan development and language

1171

attrition. In B. Köpke, M. S. Schmid, M. Keijzer, & S. Dostert (Eds.), Language

1172

Attrition: Theoretical Perspectives (pp. 53–68). Amsterdam, The Netherlands:

1173

John Benjamins Publishing.

1174

de Bot, K., Gommans, P., & Rossing, C. (1991). L1 loss in an L2 environment:

1175

Dutch immigrants in France. In H. W. Seliger & R. M. Vago (Eds.), First

1176

Language Attrition (pp. 87–98). Cambridge, UK: Cambridge University Press.

1177

de Leeuw, E. (in press). Phonetic L1 attrition. In M. S. Schmid & B. Köpke (Eds.),

1178

The Oxford Handbook of Language Attrition. Oxford, UK: Oxford University

1179

Press.

1180

de Leeuw, E., Mennen, I., & Scobbie, J. M. (2013). Dynamic systems, matura-

1181

tional constraints and L1 phonetic attrition. International Journal of Bilingual-

1182

ism, 17(6), 683–700.

58

1183

de Leeuw, E., Schmid, M. S., & Mennen, I. (2010). The effects of contact on na-

1184

tive language pronunciation in an L2 migrant setting. Bilingualism: Language

1185

and Cognition, 13(1), 33–40.

1186

de Leeuw, E., Tusha, A., & Schmid, M. S. (2018). Individual phonological attri-

1187

tion in Albanian-English late bilinguals. Bilingualism: Language and Cogni-

1188

tion, 21(2), 278–295.

1189

Dmitrieva, O., Jongman, A., & Sereno, J. (2010). Phonological neutralization by

1190

native and non-native speakers: The case of Russian final devoicing. Journal of

1191

Phonetics, 38(3), 483–492.

1192

Dussias, P. E. & Sagarra, N. (2007). The effect of exposure on syntactic parsing

1193

in Spanish–English bilinguals. Bilingualism: Language and Cognition, 10(1),

1194

101–116.

1195

Flege, J. E. (1987). The production of “new” and “similar” phones in a foreign

1196

language: Evidence for the effect of equivalence classification. Journal of Pho-

1197

netics, 15(1), 47–65.

1198

Flege, J. E. (1995). Second language speech learning: Theory, findings, and prob-

1199

lems. In W. Strange (Ed.), Speech Perception and Linguistic Experience: Issues

1200

in Cross-Language Research (pp. 233–272). Baltimore, MD: York Press.

1201

Flege, J. E. (1996). English vowel productions by Dutch talkers: More evidence

1202

for the “similar” vs “new” distinction. In A. James & J. Leather (Eds.), Second-

1203

Language Speech: Structure and Process (pp. 11–52). Berlin, Germany: Mou-

1204

ton de Gruyter.

59

1205

Flege, J. E. (2002). Interactions between the native and second-language phonetic

1206

systems. In P. Burmeister, T. Piske, & A. Rohde (Eds.), An Integrated View

1207

of Language Development: Papers in Honor of Henning Wode (pp. 217–244).

1208

Trier, Germany: Wissenschaftlicher Verlag.

1209

Flege, J. E. (2007). Language contact in bilingualism: Phonetic system interac-

1210

tions. In J. Cole & J. I. Hualde (Eds.), Laboratory Phonology 9 (pp. 353–382).

1211

Berlin, Germany: Walter de Gruyter.

1212

1213

Forster,

J.

C.

(2014).

DMDX.

Version

5.1.0.0.

http://www.u.arizona.edu/∼kforster/dmdx/dmdx.htm.

1214

Fowler, C. A., Sramko, V., Ostry, D. J., Rowland, S. A., & Hallé, P. (2008). Cross

1215

language phonetic influences on the speech of French–English bilinguals. Jour-

1216

nal of Phonetics, 36(4), 649–663.

1217

Fox, E. (1996). Cross-language priming from ignored words: Evidence for a com-

1218

mon representational system in bilinguals. Journal of Memory and Language,

1219

35(3), 353–370.

1220

Fox, J., Weisberg, S., Price, B., Adler, D., Bates, D., Baud-Bovy, G., Bolker,

1221

B., Ellison, S., Firth, D., Friendly, M., Gorjanc, G., Graves, S., Heiberger, R.,

1222

Laboissiere, R., Maechler, M., Monette, G., Murdoch, D., Nilsson, H., Ogle,

1223

D., Ripley, B., Venables, W., Walker, S., Winsemius, D., Zeileis, A., & R Core

1224

Team. (2018). car: Companion to applied regression. R package version 3.0-2.

1225

https://cran.r-project.org/web/packages/car/. Last retrieved February 20, 2019.

1226

Freedman, S. E. & Barlow, J. A. (2012). Using whole-word production measures

1227

to determine the influence of phonotactic probability and neighborhood density 60

1228

on bilingual speech production. International Journal of Bilingualism, 16(4),

1229

369–387.

1230

1231

1232

1233

1234

1235

Genesee, F. (1989). Early bilingual development: One language or two? Journal of Child Language, 16(1), 161–179. Grosjean, F. (1985). The bilingual as a competent but specific speaker-hearer. Journal of Multilingual and Multicultural Development, 6(6), 467–477. Grosjean, F. (1989). Neurolinguists, beware! The bilingual is not two monolinguals in one person. Brain and Language, 36(1), 3–15.

1236

Guion, S. G. (2003). The vowel systems of Quichua-Spanish bilinguals: Age of

1237

acquisition effects on the mutual influence of the first and second languages.

1238

Phonetica, 60(2), 98–128.

1239

1240

Guion, S. G., Flege, J. E., & Loftin, J. D. (2000). The effect of L1 use on pronunciation in Quichua-Spanish bilinguals. Journal of Phonetics, 28(1), 27–42.

1241

Hagiwara, R. (1997). Dialect variation and formant frequency: The American

1242

English vowels revisited. Journal of the Acoustical Society of America, 102(1),

1243

655–658.

1244

Harrington, J., Palethorpe, S., & Watson, C. I. (2000). Monophthongal vowel

1245

changes in Received Pronunciation: An acoustic analysis of the Queen’s Christ-

1246

mas broadcasts. Journal of the International Phonetic Association, 30(1–2),

1247

63–78.

1248

1249

Helgason, P. & Ringen, C. (2008). Voicing and aspiration in Swedish stops. Journal of Phonetics, 36(4), 607–628. 61

1250

Herd, W. J., Walden, R. L., Knight, W. L., & Alexander, S. N. (2015). Phonetic

1251

drift in a first language dominant environment. Proceedings of Meetings on

1252

Acoustics, 23, 060005.

1253

Higby, E. & Obler, L. (2014). The adaptation of native language construal patterns

1254

in second language acquisition. AFinLA-e: Soveltavan kielitieteen tutkimuksia,

1255

6, 32–44.

1256

Hillenbrand, J., Getty, L. A., Clark, M. J., & Wheeler, K. (1995). Acoustic char-

1257

acteristics of American English vowels. Journal of the Acoustical Society of

1258

America, 97(5), 3099–3111.

1259

1260

Holliday, J. J. (2015). A longitudinal study of the second language acquisition of a three-way stop contrast. Journal of Phonetics, 50, 1–14.

1261

Huffman, M. K. & Schuhmann, K. (2015). Individual differences in phonetic drift

1262

by English-speaking learners of Spanish. Journal of the Acoustical Society of

1263

America, 137(4), 2268.

1264

Jarvis, S. (2003). Probing the effects of the L2 on the L1: A case study. In V. Cook

1265

(Ed.), Effects of the Second Language on the First (pp. 81–102). Clevedon, UK:

1266

Multilingual Matters.

1267

1268

1269

1270

Jarvis, S. & Pavlenko, A. (2008). Crosslinguistic Influence in Language and Cognition. New York, NY: Routledge. Jessner, U. (1999). Metalinguistic awareness in multilinguals: Cognitive aspects of third language learning. Language Awareness, 8(3–4), 201–209.

62

1271

Joh, J., Ko, H., Lim, I., & Lee, S. (2010). The impact of English language learn-

1272

ing on the acquisition of native language among Korean children. Eungyong

1273

Eoneohag [Korean Journal of Applied Linguistics], 26(3), 101–141.

1274

1275

Kartushina, N. (2015). Second Language Phonological Acquisition in Adults. PhD thesis, University of Geneva, Geneva, Switzerland.

1276

Kartushina, N., Frauenfelder, U. H., & Golestani, N. (2016). How and when

1277

does the second language influence the production of native speech sounds: A

1278

literature review. Language Learning, 66(S2), 155–186.

1279

Kartushina, N., Hervais-Adelman, A., Frauenfelder, U. H., & Golestani, N.

1280

(2016). Mutual influences between native and non-native vowels in produc-

1281

tion: Evidence from short-term visual articulatory feedback training. Journal

1282

of Phonetics, 57, 21–39.

1283

1284

1285

1286

1287

1288

1289

1290

1291

Keating, P., Linker, W., & Huffman, M. (1983). Patterns in allophone distribution for voiced and voiceless stops. Journal of Phonetics, 11(3), 277–290. Keating, P. A. (1984). Phonetic and phonological representation of stop consonant voicing. Language, 60(2), 286–319. Kecskes, I. (1998). The state of L1 knowledge in foreign language learners. Word, 49(3), 321–340. Kecskes, I. & Papp, T. (2000). Foreign Language and Mother Tongue. Mahwah, NJ: Lawrence Erlbaum Associates. Kim, J.-H., Montrul, S., & Yoon, J. (2010). Dominant language influence in

63

1292

acquisition and attrition of binding: Interpretation of the Korean reflexive caki.

1293

Bilingualism: Language and Cognition, 13(1), 73–84.

1294

Kim, K. H. S., Relkin, N. R., Lee, K.-M., & Hirsch, J. (1997). Distinct cortical

1295

areas associated with native and second languages. Nature, 388(6638), 171–

1296

174.

1297

Kim, M. (2009). Phonetic accommodation in conversations between native and

1298

nonnative speakers. Journal of the Acoustical Society of America, 125(4), 2764.

1299

Kim, M., Horton, W. S., & Bradlow, A. R. (2011). Phonetic convergence in spon-

1300

taneous conversations as a function of interlocutor language distance. Labora-

1301

tory Phonology, 2(1), 125–156.

1302

Kroll, J. F., Bobb, S. C., & Hoshino, N. (2014). Two languages in mind: Bilin-

1303

gualism as a tool to investigate language, cognition, and the brain. Current

1304

Directions in Psychological Science, 23(3), 159–163.

1305

1306

Ladefoged, P. (2005). Vowels and Consonants (2nd ed.). Malden, MA: Blackwell Publishing.

1307

Lai, Y., Huff, C., Sereno, J., & Jongman, A. (2009). The raising effect of aspi-

1308

rated prevocalic consonants on f0 in Taiwanese. In Brooke, J., Coppola, G.,

1309

Görgülü, E., Mameni, M., Mileva, E., Morton, S., & Rimrott, A. (Eds.), Pro-

1310

ceedings of the 2nd International Conference on East Asian Linguistics, Van-

1311

couver, Canada. Simon Fraser University.

1312

Lang, B. & Davidson, L. (in press). Effects of exposure and vowel space distri-

1313

bution on phonetic drift: Evidence from American English learners of French.

1314

Language and Speech. 64

1315

Laufer, B. & Baladzhaeva, L. (2015). First language attrition without second

1316

language acquisition: An exploratory study. International Journal of Applied

1317

Linguistics, 166(2), 229–253.

1318

Lewis, P. M., Simons, G. F., & Fennig, C. D. (2015). Ethnologue: Languages of

1319

the world. 18th edition. http://www.ethnologue.com. Last retrieved March 8,

1320

2015.

1321

Li, P., Zhang, F., Tsai, E., & Puls, B. (2014). Language history questionnaire

1322

(LHQ 2.0): A new dynamic web-based research tool. Bilingualism: Language

1323

and Cognition, 17(3), 673–680.

1324

Linck, J. A., Kroll, J. F., & Sunderman, G. (2009). Losing access to the native lan-

1325

guage while immersed in a second language: Evidence for the role of inhibition

1326

in second-language learning. Psychological Science, 20(12), 1507–1515.

1327

1328

Lisker, L. & Abramson, A. S. (1964). A cross-language study of voicing in initial stops: Acoustical measurements. Word, 20(3), 384–422.

1329

Lisker, L., Liberman, A. M., Erickson, D. M., Dechovitz, D., & Mandler, R.

1330

(1977). On pushing the voice-onset-time (VOT) boundary about. Language

1331

and Speech, 20(3), 209–216.

1332

1333

1334

1335

Lüpke, F. & Storch, A. (2013). Repertoires and Choices in African Languages. Boston, MA: Walter de Gruyter. Major, R. C. (1992). Losing English as a first language. The Modern Language Journal, 76(2), 190–208.

65

1336

Marian, V., Blumenfeld, H. K., & Kaushanskaya, M. (2007). The Language Ex-

1337

perience and Proficiency Questionnaire (LEAP-Q): Assessing language profiles

1338

in bilinguals and multilinguals. Journal of Speech, Language, and Hearing Re-

1339

search, 50(4), 940–967.

1340

Marian, V., Spivey, M., & Hirsch, J. (2003). Shared and separate systems in bilin-

1341

gual language processing: Converging evidence from eyetracking and brain

1342

imaging. Brain and Language, 86(1), 70–82.

1343

Mayr, R., Price, S., & Mennen, I. (2012). First language attrition in the speech of

1344

Dutch–English bilinguals: The case of monozygotic twin sisters. Bilingualism:

1345

Language and Cognition, 15(4), 687–700.

1346

1347

Mennen, I. (2004). Bi-directional interference in the intonation of Dutch speakers of Greek. Journal of Phonetics, 32(4), 543–563.

1348

Morris, R. J., McCrea, C. R., & Herring, K. D. (2008). Voice onset time differ-

1349

ences between adult males and females: Isolated syllables. Journal of Phonet-

1350

ics, 36(2), 308–317.

1351

Namjoshi, J., Tremblay, A., Spinelli, E., Broersma, M., Martínez-García, M. T.,

1352

Connell, K., Cho, T., & Kim, S. (2015). Speech segmentation is adaptive even

1353

in adulthood: Role of the linguistic environment. In The Scottish Consortium

1354

for ICPhS 2015 (Ed.), Proceedings of the 18th International Congress of Pho-

1355

netic Sciences (pp. paper number 0676). Glasgow, UK: University of Glasgow.

1356

Nearey, T. M. & Rochet, B. L. (1994). Effects of place of articulation and vowel

1357

context on VOT production and perception for French and English stops. Jour-

1358

nal of the International Phonetic Association, 24(1), 1–18. 66

1359

Oh, G. E., Guion-Anderson, S., Aoyama, K., Flege, J. E., Akahane-Yamada, R.,

1360

& Yamada, T. (2011). A one-year longitudinal study of English and Japanese

1361

vowel production by Japanese adults and children in an English-speaking set-

1362

ting. Journal of Phonetics, 39(2), 156–167.

1363

Oh, J. S., Au, T. K., & Jun, S.-A. (2010). Early childhood language memory in the

1364

speech perception of international adoptees. Journal of Child Language, 37(5),

1365

1123–1132.

1366

1367

Paradis, J. (2001). Do bilingual two-year-olds have separate phonological systems? International Journal of Bilingualism, 5(1), 19–38.

1368

Parmentier, F. B. R., Elsley, J. V., & Ljungberg, J. K. (2010). Behavioral distrac-

1369

tion by auditory novelty is not only about novelty: The role of the distracter’s

1370

informational value. Cognition, 115(3), 504–511.

1371

1372

Pavlenko, A. (2000). L2 influence on L1 in late bilingualism. Issues in Applied Linguistics, 11(2), 175–205.

1373

Pavlenko, A. (2003). ‘I feel clumsy speaking Russian’: L2 influence on L1 in nar-

1374

ratives of Russian L2 users of English. In V. Cook (Ed.), Effects of the Second

1375

Language on the First (pp. 32–61). Clevedon, UK: Multilingual Matters.

1376

1377

1378

1379

Pavlenko, A. & Jarvis, S. (2002). Bidirectional transfer. Applied Linguistics, 23(2), 190–214. Peterson, G. E. & Barney, H. L. (1952). Control methods used in a study of the vowels. Journal of the Acoustical Society of America, 24(1), 175–184.

67

1380

1381

Piller, I. (2002). Passing for a native speaker: Identity and success in second language learning. Journal of Sociolinguistics, 6(2), 179–206.

1382

R Development Core Team (2018). R: A Language and Environment for Sta-

1383

tistical Computing. Vienna, Austria: R Foundation for Statistical Computing.

1384

http://www.r-project.org. Last retrieved February 20, 2019.

1385

1386

1387

1388

1389

1390

1391

1392

Rickford, J. & Price, M. (2013). Girlz II women: Age-grading, language change and stylistic variation. Journal of Sociolinguistics, 17(2), 143–179. Sancier, M. L. & Fowler, C. A. (1997). Gestural drift in a bilingual speaker of Brazilian Portuguese and English. Journal of Phonetics, 27(4), 421–436. Sankoff, G. & Blondeau, H. (2007). Language change across the lifespan: /r/ in Montreal French. Language, 83(3), 560–588. Schmid, M. S. (2013). First language attrition. Wiley Interdisciplinary Reviews: Cognitive Science, 4(2), 117–123.

1393

Schmid, M. S. & Köpke, B. (2009). L1 attrition and the mental lexicon. In

1394

A. Pavlenko (Ed.), The Bilingual Mental Lexicon: Interdisciplinary Approaches

1395

(pp. 209–238). Bristol, UK: Multilingual Matters.

1396

Schuhmann, K. S. & Huffman, M. K. (2015). L1 drift and L2 category formation

1397

in second language learning. In The Scottish Consortium for ICPhS 2015 (Ed.),

1398

Proceedings of the 18th International Congress of Phonetic Sciences (pp. paper

1399

number 0850). Glasgow, UK: University of Glasgow.

1400

Schwanenflugel, P. J. & Rey, M. (1986). Interlingual semantic facilitation: Evi-

68

1401

dence for a common representational system in the bilingual lexicon. Journal

1402

of Memory and Language, 25(5), 605–618.

1403

Seliger, H. W. & Vago, R. M. (1991). The study of first language attrition: An

1404

overview. In H. W. Seliger & R. M. Vago (Eds.), First Language Attrition (pp.

1405

3–15). Cambridge, UK: Cambridge University Press.

1406

Sharma, D. & Sankaran, L. (2011). Cognitive and social forces in dialect shift:

1407

Gradual change in London Asian speech. Language Variation and Change,

1408

23(3), 399–428.

1409

Stolberg, D. & Münch, A. (2010). “Die Muttersprache vergisst man nicht” – or

1410

do you? A case study in L1 attrition and its (partial) reversal. Bilingualism:

1411

Language and Cognition, 13(1), 19–31.

1412

Tice, M. & Woodley, M. (2012). Paguettes and bastries: Novice French learners

1413

show shifts in native phoneme boundaries. UC Berkeley Phonology Lab Annual

1414

Report, 2012, 72–75.

1415

Tobin, S. J., Nam, H., & Fowler, C. A. (2017). Phonetic drift in Spanish-English

1416

bilinguals: Experiment and a self-organizing model. Journal of Phonetics, 65,

1417

45–59.

1418

Tsimpli, I., Sorace, A., Heycock, C., & Filiaci, F. (2004). First language attrition

1419

and syntactic subjects: A study of Greek and Italian near-native speakers of

1420

English. International Journal of Bilingualism, 8(3), 257–277.

1421

Tucker, G. R. (1999). A global perspective on bilingualism and bilingual educa-

1422

tion. ERIC Digest EDO-FL-99-04, Center for Applied Linguistics, Washington,

1423

DC. 69

1424

Ulbrich, C. & Ordin, M. (2014). Can L2-English influence L1-German? The case

1425

of post-vocalic /r/. Journal of Phonetics, 45, 26–42.

1426

Unsworth, S. (2013). Assessing the role of current and

CUMULATIVE

exposure

1427

in simultaneous bilingual acquisition: The case of Dutch gender. Bilingualism:

1428

Language and Cognition, 16(1), 86–110.

1429

van Hell, J. G. & Dijkstra, T. (2002). Foreign language knowledge can influ-

1430

ence native language performance in exclusively native contexts. Psychonomic

1431

Bulletin and Review, 9(4), 780–789.

1432

1433

1434

1435

Wagner, S. E. (2012). Real-time evidence for age grad(ing) in late adolescence. Language Variation and Change, 24(2), 179–202. Wagner, S. E. & Sankoff, G. (2011). Age grading in the Montréal French inflected future. Language Variation and Change, 23(3), 275–313.

1436

Ward, N., Sundara, M., Conboy, B., & Kuhl, P. (2009). Consequences of short-

1437

term language exposure in infancy on babbling. Journal of the Acoustical So-

1438

ciety of America, 126(4), 2311.

1439

1440

Yang, B. (1996). A comparative study of American English and Korean vowels produced by male and female speakers. Journal of Phonetics, 24(2), 245–261.

1441

Yelland, G. W., Pollard, J., & Mercuri, A. (1993). The metalinguistic benefits

1442

of limited contact with a second language. Applied Psycholinguistics, 14(4),

1443

423–444.

1444

Yeni-Komshian, G. H., Flege, J. E., & Liu, S. (2000). Pronunciation proficiency

70

1445

in the first and second languages of Korean-English bilinguals. Bilingualism:

1446

Language and Cognition, 3(2), 131–149.

1447

Yoon, S.-Y. (2015). Acoustic properties of Korean stops as L1 produced by L2

1448

learners of the English language. Communication Sciences and Disorders,

1449

20(2), 178–188.

71

View publication stats